Traditional non-destructive testing methods for industrial assets require sending inspectors to elevated and confined areas using scaffolding or rope access. This method is expensive, time-consuming, and dangerous.Avestec has developed a solution using intelligent robots that fly, attach, and measure. Completing all inspection requirements quickly, safely, and at a low cost.With the technology of intelligent flying robots, what once took an entire inspection team for ten days can be completed by two operators in a single day.Unlimited flight and a patented attachment mechanism enable our robots to perform Ultrasonic Thickness measurements and visual inspections easily.A cloud-based reporting software like Avesoft handles all the post-processing and data analysis using artificial intelligence and automatically generates reports. So well-informed decisions can be made faster.

Intelligent flying robots are making their way into the Non-Destructive Testing industry.
These robots are gaining attention in providing the industry with a safer and superior alternative to traditional methods.

Flying into Non-Destructive Testing

Another fascinating month for robotics. Researchers are the highlight of the month. Through their work, we are pushing the boundaries of self-awareness, complex structure and even the limits of life and death.&nbsp;Spoooky indeed.&nbsp;Take a seat and join us in this quick review of robotics news.And don&#39;t forget that you can also share your story. We would love to feature it in our next blog. Just email us at <a href="mailto:robotics.community@canonical.com" target="_blank">robotics.community@canonical.com</a><ul><li>This blog has also been <a href="http://ubuntu.com/blog/the-state-of-robotics-august-2022?utm_source=wevolver&utm_medium=blog" rel="noopener noreferrer" target="_blank">posted in Ubuntu blogs</a>.&nbsp;</li></ul><h2 dir="ltr">Security for robotics</h2><h3 dir="ltr">Vulnerability in cute, smart Enabot Ebo Air robot found and fixed</h3>The <a href="https://na.enabot.com/ebo/AIR" target="_blank">Enabot Ebo Air</a> is a friendly smart home robot, designed to keep the family company. It allows you to interact with your family at home while you&rsquo;re away. It can take photos and videos as well as recognise faces, and you can steer it around the house via Wi-Fi. Recently, the security assessment firm Modux was asked to evaluate the security of the robot, and <a href="https://www.modux.co.uk/post/serious-security-issues-uncovered-with-the-enabot-smart-robot" target="_blank">found two important vulnerabilities</a>. Modux disclosed the issues to the manufacturer, who has since addressed the risks and released a fix.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1ODQzODQ3LXRlc3QgMS53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">The first issue found was the use of a shared password for the system administrator (root) account. The Enabot Ebo Air robot came pre-configured with a default system administrator password, which was weak and shared across all devices. An attacker could have used this to connect via SSH (secure shell), and gain full administrative permissions. The recommendation was to disable SSH functionality, which the manufacturer did.&nbsp;The second bug was a missing secure-delete functionality, which could potentially disclose information from the previous owner if the robot was resold. The robot did not effectively perform a secure deletion of data when a factory reset was initiated, leaving behind artefacts from previous uses of the device, including cleartext Wi-Fi passwords and session tokens from the previous owner. For this, the recommendation was to use a secure deletion functionality instead, overwriting all sensitive data multiple times.An attacker could have used these vulnerabilities to move the robot, access the camera, record video, speak through the microphone, among other tasks, effectively using it as a surveillance device. This event highlights the importance of security awareness and thorough testing, as robots continue to serve and accompany us more and more in our daily lives. <h2 dir="ltr">Amazon acquisition of iRobot - implications beyond data</h2>Every time Amazon launches or acquires a new company (especially one developing a device) the whole internet starts talking about the data and privacy implications. Yes, certainly data is important for Amazon. But that is not the whole story. Today I want to talk about the impact this acquisition has on investors and how this could boost the market.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1ODg4NDM2LXRlc3QyLndlYnAiLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">Funding a robotics venture is not easy. Between March 2020 and March 2021, robotics companies raised only 8% in capital compared to AI companies. AI-focused startups raised a total of $73.4 billion while robotics startups raised $6.3 billion. This is a paltry sum in comparison, especially when we saw an increasing need for robotics in several industries (due to COVID). &nbsp;&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1OTA4ODQxLXRlc3QzLndlYnAiLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">There are many reasons behind this. One of those, and the most important one, is the increased risk perception from investors. Robotics development is expensive and takes time.&nbsp;But moves like Amazon&rsquo;s show something to investors: a profitable exit. Amazon will acquire iRobot for $61 per share in an all-cash transaction valued at approximately $1.7 billion. A success story is always needed to push investment, and this certainly makes one.&nbsp;This is not an isolated event. It comes after 2 years of really interesting acquisitions. Take <a href="https://www.zebra.com/gb/en/about-zebra/newsroom/press-releases/2021/zebra-technologies-to-acquire-fetch-robotics.html" target="_blank">Zebra&#39;s acquisition of Fetch</a>, or <a href="https://www.deere.com/en/news/all-news/2021aug5-bear-flag-robotics/" target="_blank">John Deere&#39;s acquisitions of ROS-based startups</a>. Investors are aware, now more than ever, that while the risk is high, the profit/ROI could be as well. Moves like <a href="https://ubuntu.com/blog/the-state-of-robotics-september-2021" target="_blank">Amazon launching Astro</a> create more incentives.&nbsp;Amazon&rsquo;s acquisition has various implications. Only time will tell whether this boosts the market, and how much.&nbsp;<h2 dir="ltr">Robotics awareness&nbsp;</h2>Robots are now able to learn a model of their entire body from scratch, without any human assistance. <a href="https://www.science.org/doi/10.1126/scirobotics.abn1944" target="_blank">Researchers from Columbia University</a> developed a visual self-model using deep neural networks that allowed a robot to learn the relationship between its motor actions and the volume it occupied in its environment.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1OTM2ODEyLXRlc3QgNC53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">The team found a way to visualise how the robot saw itself, finding a surprisingly accurate self-image. (<a href="https://www.hackster.io/news/columbia-university-researchers-gave-this-robot-arm-a-true-sense-of-self-cadeac6481a9" target="_blank">source</a>) The researchers placed a robotic arm inside a circle of five streaming video cameras. The video footage was fed to the network as the robot wiggled and contorted. This allowed it to learn exactly how its body moved in response to various motor commands.&nbsp;After about three hours, the robot learned the relationship between its motor actions and the volume it occupied in its environment. Much like an infant exploring itself for the first time in a hall of mirrors.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1OTU2Njk5LXRlc3Q1LndlYnAiLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">Robot inside a circle of five streaming video cameras This awareness allows the robot to plan motion, reach goals, and avoid obstacles in a variety of situations&mdash;another small but incredible step towards robots&rsquo; self-awareness. <h2 dir="ltr">Robotics and Smart textiles</h2><a href="https://newsroom.unsw.edu.au/news/science-tech/new-shape-shifting-material-can-move-robot" target="_blank">Researchers at the University of New South Wales</a> are also working on a special project. They developed a new class of smart textiles that can shape-shift and turn two-dimensional material into 3D structures. The material is constructed from tiny soft artificial muscles. These muscles are long silicone tubes that can change shape thanks to hydraulics.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU1OTc3MTkxLXRlc3Q2LndlYnAiLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">The UNSW team&#39;s smart textile enables fabric reconfiguration which produces shape-morphing structures such as this butterfly and flower that can move using hydraulics. (<a href="https://newsroom.unsw.edu.au/news/science-tech/new-shape-shifting-material-can-move-robot" target="_blank">source</a>) The artificial muscles can be programmed to contract or expand into a variety of shapes depending on their initial structure. The smart textile can either be attached to existing passive material or yarn to create an active fabric.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU2MDAwMzM2LXRlc3QgNy53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">Structure of smart textile (<a href="https://newsroom.unsw.edu.au/news/science-tech/new-shape-shifting-material-can-move-robot" target="_blank">source</a>)&nbsp;&nbsp;<h2 dir="ltr">Spider gripper&nbsp;</h2>What&rsquo;s next is not fabric, but something from a sci-fi movie. <a href="https://news.rice.edu/news/2022/rice-engineers-get-grip-necrobotic-spiders" target="_blank">Rice University researchers</a> found a way to use dead spiders as robotic grippers. What started with a dead spider in a lab became a gripper; Necrobotics.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU2MDI3MDc0LXRlc3QgOC53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">Functioning of the spider gripper (<a href="https://www.gizmodo.com.au/2022/07/necrobotic-spiders/" target="_blank">source</a>) If you haven&#39;t noticed yet, spiders&rsquo; legs curl up when they die. This is because spiders extend their legs with hydraulic pressure and when spiders die they lose this ability. Researchers tapped into the hydraulic infrastructure of dead spiders using a needle, and by pumping air, they extended the spider&rsquo;s legs. Releasing the pressure retracted the legs, allowing them to grab objects. <h2 dir="ltr">Congratulations to ROS Industrial&nbsp;for its 10-year anniversary</h2>Happy birthday <a href="https://rosindustrial.org/" target="_blank">ROS Industrial</a>! If you don&#39;t know it yet, ROS Industrial is an open source project that seeks to extend ROS and now ROS 2 to industrial hardware and applications.&nbsp;ROS Industrial identifies and prioritises capabilities for industrial robotics and automation to address current and future application problems. They also provide a wide range of user services, including technical support and training, to facilitate the continued adoption of ROS by the industry.<iframe width="640" height="360" src="https://www.youtube.com/embed/-6yAk05et1Q?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" data-gtm-yt-inspected-12="true" id="26562603" title="ROS-Industrial 10 Year Anniversary Mash Up"></iframe> They help us all carry the flag of open source in the industrial and automation domain. For that, we are truly thankful and we&rsquo;re proud to be part of this community. &nbsp;<h2 dir="ltr" id="isPasted">Robot Makers - Chapter 3 - Register now!&nbsp;</h2>Did you miss the interview with Akara Robotics CEO Conor McGinn? Don&rsquo;t worry,<a href="https://www.youtube.com/watch?v=4O3X7MM1mx8" target="_blank">&nbsp;the video is now available on youtube</a>.&nbsp;And don&#39;t miss our <a href="https://www.brighttalk.com/webcast/6793/555737?utm_source=Canonical&utm_medium=brighttalk&utm_campaign=555737?utm_source=wevolver&utm_medium=engage" rel="noopener noreferrer" target="_blank">upcoming interview with Russell Nickerson</a>, Engagement Liaison at MassRobotics. MassRobotics is a non-profit coworking space for robotics startups in Boston&#39;s Seaport, where Russell supports more than 60 residence startups.&nbsp;Based on his experience scaling robotics companies, Russell will share the dos and don&#39;ts of go-to-market strategy.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYxMTYwMTQ5MDQyLVtXZWJpbmFyXSBNYXNzUm9ib3RpY3MgUm9ib3QgTWFrZXJzIDIgKDIpLnBuZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" class="fr-fic fr-dii"><ul><li dir="ltr">Register now for <a href="https://www.brighttalk.com/webcast/6793/555737?utm_source=Canonical&utm_medium=brighttalk&utm_campaign=555737?utm_source=wevolver&utm_medium=engage" rel="noopener noreferrer" target="_blank">Chapter 3 of the Robot Maker series</a></li></ul><h2 dir="ltr">Open source robotics &ndash; white papers</h2>We recently published a whitepaper about the importance of<a href="https://ubuntu.com/engage/robot-operating-system-choice?utm_source=wevolver&utm_medium=engage" rel="noopener noreferrer" target="_blank">&nbsp;choosing the right OS for your robot</a>. Picking the right OS could make a big difference in factors like security, maintenance and infrastructure for deployment.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU2MDcxNzI2LXRlc3QgOS53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib"><ul><li>Review the implications and explore <a href="https://ubuntu.com/engage/robot-operating-system-choice?utm_source=wevolver&utm_medium=engage" id="isPasted" rel="noopener noreferrer" target="_blank">why so many developers chose Ubuntu</a>.&nbsp;</li></ul><h2 dir="ltr">Open source robotics &ndash; video news</h2>Another month, another video. More news, more applications, more robots! &nbsp;<iframe width="640" height="360" src="https://www.youtube.com/embed/O29XoRFNc0k?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe> If you want us to feature your robotics application, just send us an email at <a data-fr-linked="true" href="mailto:robotics.community@canonical.com" rel="noopener noreferrer" target="_blank">robotics.community@canonical.com</a>.<ul><li dir="ltr">And don&#39;t forget to <a href="https://www.youtube.com/c/UbuntuOS/?sub_confirmation=1" rel="noopener noreferrer" target="_blank">follow us for more robotics news, demos and tutorials!</a>&nbsp;</li></ul><h2 dir="ltr">Open-source robotics &ndash; ROS 2 shared memory tutorial</h2>After receiving questions regarding an error message issued by ROS 2 applications in snaps, we decided to put together a blog post on this matter <a href="https://ubuntu.com/blog/how-to-use-ros-2-shared-memory-in-snaps?utm_source=wevolver&utm_medium=blog" rel="noopener noreferrer" target="_blank">&ldquo;How to use ROS 2 shared memory in snaps&rdquo;</a><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwNTU2MDkzNDI3LXRlc3QgMTAud2VicCIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 300px;" class="fr-fic fr-dib"><a href="https://ubuntu.com/blog/how-to-use-ros-2-shared-memory-in-snaps?utm_source=wevolver&utm_medium=blog" rel="noopener noreferrer" target="_blank">In this post</a>, we review the shared memory mechanism in ROS 2, explain why the shared memory feature causes error messages in snaps and propose different solutions to tackle it.As usual, we&rsquo;d love to get your feedback. Feel free to reach out on the <a href="https://forum.snapcraft.io/" target="_blank">snapcraft forum</a>.<h2 dir="ltr">Stay tuned for more robotics news</h2>As always, we would love to hear from you. Send a summary of your robotics innovation and project to <a href="mailto:robotics.community@canonical.com" target="_blank">robotics.community@canonical.com</a>, and we will share it in our next robotics newsletter or monthly video. Thanks for reading!

Monthly robotics news blog. Discover the latest innovations, products, research, tutorials, and security news for robotics.

The State of Robotics, August 2022

In this episode, we talk about how engineers are tackling the problem of sorting through piles using robotics and a platform that leverages fake data to train robots more efficiently.<iframe height="200px" width="100%" frameborder="no" scrolling="no" seamless="" src="https://player.simplecast.com/ee270fa2-ec1d-485d-86fe-fffdf942a9f3?dark=false" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe><hr>This podcast is sponsored by&nbsp;<a href="https://www2.mouser.com/" id="isPasted" rel="nofollow" target="_blank">Mouser Electronics</a>. &nbsp;<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1658227935701-Mouser-sponsor-1-a.jpg" style="width: 766px;" class="fr-fic fr-dib"><hr><h3 id="isPasted">EPISODE NOTES</h3>(0:57) -&nbsp;<a href="https://www.wevolver.com/article/robot-overcomes-uncertainty-to-retrieve-buried-objects" target="_blank">Robots Finding Buried Objects🤖🔑</a>(16:50) - <a href="https://www.wevolver.com/article/fake-data-helps-robots-learn-the-ropes-faster" target="_blank">Fake Data To Train Robots🤖🧠</a><hr>About the podcast:Every day, some of the most innovative universities, companies, and individual technology developers share their knowledge on Wevolver. To ensure we can also provide this knowledge for the growing group of podcast listeners, we started a collaboration with two young engineers, Daniel Scott Mitchell &amp; Farbod Moghaddam who discuss the most interesting content in this podcast series.&nbsp;To learn more about this show, please visit the <a href="https://www.wevolver.com/profile/the.next.byte" target="_blank">shows page</a>. By following the page, you will get automatic updates by email when a new show is published.Be sure to give us a follow and review on <a href="https://podcasts.apple.com/nl/podcast/the-next-byte/id1548255861?l=en" rel="nofollow" target="_blank">Apple</a> podcasts, <a href="https://open.spotify.com/show/6lPPsRtegnivsRUEsQ09R7?si=IPMiah7xT_apJtPjqvXMlA" rel="nofollow" target="_blank">Spotify</a>, and most of your favorite podcast platforms!Take a few seconds to leave us a review. It really helps! <a href="https://apple.co/2RIsbZ2" rel="nofollow" target="_blank">https://apple.co/2RIsbZ2&nbsp;</a>if you do it and send us proof, we&rsquo;ll give you a shoutout on the show.

Peter Mitrano demonstrates a rope manipulation experiment. Software that he and Berenson developed can expand training data sets for challenges like rope manipulation, doubling the success rate of the robot. Credit: Daryl Marshke, Michigan Creative, University of Michigan. 

In this episode, we talk about how engineers are tackling the problem of sorting through piles using robotics and a platform that leverages fake data to train robots more efficiently.

Podcast: Teaching Robots To Find Buried Objects

<h2 dir="ltr" id="isPasted">The State of Robotics &ndash; May &amp; June 2022</h2>In May, robotics felt the waves of the Hawksbill and Jellyfish. Last month brought us the first LTS for ROS 2, Humble Hawksbill. As part of the ROS community, we welcome this new release and congratulate everyone who made it possible.In this round-up, we will talk about ROS, drones, football and robots that eat scrambled eggs.&nbsp;Yep, enjoy ;) &nbsp;&nbsp;And don&#39;t forget that you can also share your story. We would love to feature it in our next blog. Just email us at <a href="mailto:robotics.community@canonical.com" target="_blank">robotics.community@canonical.com</a>&nbsp;<h2 dir="ltr">Security for robotics</h2><h3 dir="ltr">Critical vulnerabilities found in hospital robots</h3>Five critical bugs were <a href="https://technewsmax.com/2022/04/12/critical_vuln_hospital_robots/" target="_blank">recently found</a> in TUG autonomous mobile robots. These internet-connected robots are being used at hundreds of hospitals around the world to deliver medicine and maintenance supplies.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI4OTAxODY0LXR1Z2d5LWlubGluZS53ZWJwIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">The zero-day vulnerabilities (CVSS scores ranging from 7.6 to 9.8), collectively called <a href="https://www.cynerio.com/jekyllbot-5-command-center" target="_blank">JekyllBot:5</a>, required a low skill set and could be exploited without any special privileges or user interaction. When exploited, they enabled remote attacker control of the robots and access to real-time camera feeds and device data. The vulnerabilities originated in TUG Homebase Server&#39;s JavaScript and API implementation, and a WebSocket that relied on absolute trust between the server and the robots to relay commands to them. Some HTML vulnerabilities on the web portal page also allowed an attacker to insert malicious Javascript code on any computer that requested data about the robots. Cynerio, the threat research firm that discovered the bugs, immediately notified hospitals using the robot.&nbsp;&nbsp;Attacks on hospital robots could have critical consequences that impact patient care, such as disrupting or blocking drug delivery or even causing damage by crashing them into people or objects. Thankfully, none of these vulnerabilities were exploited in the wild. All five bugs have now been fixed for all TUG base server versions before version 24.<h2 dir="ltr">ROS 2 Humble, Ubuntu 22.04 and Ubuntu Core 22</h2>With the release of Ubuntu 22.04, we have a new release of ROS 2 Humble Hawksbill. This long-term support (LTS) release is the first LTS for a ROS 2 distribution. These 5 years of support show the maturity that ROS 2 has achieved. As such, it brings <a href="https://discourse.ros.org/t/ros-2-humble-hawksbill-released/25729" target="_blank">several features</a>:<ul><li dir="ltr">Type adaptation: In rclcpp, you can work with custom data structures and have them implicitly converted to be sent over the ROS network, or passed directly if you&rsquo;re using interprocess communication. This is convenient for working with complex data structures like images.</li><li dir="ltr">launch_pytest: A pytest plugin for writing tests in Python.</li><li dir="ltr">Frontend support for composable nodes: Composable nodes can now be specified in launch files.</li><li dir="ltr">Content filtered topics: Subscribers can filter which types of messages they want to see.</li></ul><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI4OTk1NTkyLXJvcy1odW1ibGUtaGF3a3NiaWxsLWZlYXR1cmVkLndlYnAiLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">And what better match for Humble Hawksbill than Jammy Jellyfish. Ubuntu 22.04 is now generally available, featuring significant leaps forward in cloud confidential computing, real-time kernel for industrial applications, and enterprise Active Directory, PCI-DSS, HIPAA, FIPS and FedRAMP compliance. This release raises the bar for open source from cloud to edge, IoT and workstations.&nbsp;For robotics development these are the most attractive features in this new distribution:<ul><li dir="ltr">Real-time kernel for Ubuntu, with the PREEMPT_RT patchset integration. This enables extreme latency-dependent use cases, providing deterministic response times to service events. Available for x86_64 and AArch64.</li><li dir="ltr">FIPS-certified Ubuntu images that make it easy for companies to fulfil different industrial compliance requirements (e.g. PCI-DSS, HIPAA and FedRAMP).</li><li dir="ltr">OVAL integration for vulnerability and patch definitions to enable auditing for Common Vulnerabilities and Exposures (CVEs)s.</li><li dir="ltr">Active Directory is now fully supported in the Ubuntu installer with Advanced Group Policy Object and allows more refined user permissions and script execution control from within Active Directory.&nbsp;</li></ul><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5MDQwMzc5LXVidW50dSBqZWxseWZpc2gud2VicCIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 300px;" class="fr-fic fr-dib">Together with 22.04, we also released Ubuntu Core 22. Canonical announced that <a href="http://ubuntu.com/core" target="_blank">Ubuntu Core 22</a>, the fully containerised Ubuntu 22.04 LTS variant optimised for IoT and edge devices, is now generally available for download from <a href="https://ubuntu.com/download/iot" target="_blank">ubuntu.com/download/iot</a>. Combined with Canonical&rsquo;s technology offer, this release brings Ubuntu&rsquo;s comprehensive and industry-leading operating system (OS) and services to a complete range of embedded and IoT devices.For those deploying robotics devices to market, Ubuntu Core 22 brings these new features:&nbsp;<ul><li dir="ltr">It provides a path to upgrade UC20 systems to UC22 and backward compatibility of the new features to UC20.</li><li dir="ltr">Validation set that specified what applications are either required to be installed together or are permitted to be installed together on a device or system.&nbsp;</li><li dir="ltr">Quota group sets to limit services inside applications like maximum memory and CPU usage.</li></ul>To learn more about the new features on Core 22 <a href="https://ubuntu.com/blog/what-youre-missing-out-if-you-dont-try-ubuntu-core-22" target="_blank">check the next blog</a>.&nbsp;<iframe width="640" height="360" src="https://www.youtube.com/embed/6NDWqH1SrGs?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe> <h2 dir="ltr">Skybrush going Open Source&nbsp;</h2>We stand behind open source projects since they make way for innovation. Just take a look at <a href="https://skybrush.io/" target="_blank">Skybrush</a>, a drone-show management platform. In early May, Skybrush announced the release of their ultimate multi-UAV mission and drone show management platform as an open-source project.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5MDY2Mjg4LXNreWJydXNoX2xpdmVfb25fZGVza3RvcC5qcGciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">Skybrush has also advanced on the drone swarm research front, with 30, 50, and up to 100 autonomous drones flying together in different cooperative scenarios. Besides putting on an impressive show, the drone orchestration platform can be used in other industries. It has practical applications for agriculture, surveillance and delivery - among other use cases.<h2 dir="ltr">Making drone nests with Wings</h2>Looking for another drone use case? Check <a href="https://wing.com/" target="_blank">Wing</a> and their drone delivery system. Wing &nbsp;is growing in Australia. The company allows small businesses to access a delivery infrastructure. Companies can ship their products through a system of different bases in the city, called wing nests. Once on the wing nest, drones can travel up to 6 miles in 6 minutes. This covers a wide range of houses. The delivery zones can expand as more wing nests get added to the network.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5NTQxMjQ3LUZJX2hlYWRlcl8yMDQ4eDExNTEgY29weS1taW4uanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib"><h2 dir="ltr">Drones and strong winds &nbsp;</h2>To autonomously perform delivery tasks, drones need to be able to adapt to wind conditions in real-time. This is a field that is not well understood yet. <a href="https://www.caltech.edu/about/news/rapid-adaptation-of-deep-learning-teaches-drones-to-survive-any-weather" target="_blank">A team of engineers from Caltec</a>h is expanding our knowledge in this area. They developed Neural-Fly, a deep-learning method that can help drones cope with new and unknown wind conditions in real-time by incorporating pre-trained representations through deep learning. Take note, drone fans.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5MTUzNTA5LUNhbHRlY2hfTmV1cmFsX0ZseV8wNi4yZTE2ZDBiYS5maWxsLTE2MDB4ODEwLWMxMDAuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">Neural-Fly uses an algorithm to learn the shared representation, using only 12 minutes of flight data. With the learned representation, Neural-Fly achieves precise flight control with substantially smaller tracking errors. It&rsquo;s been used to improve drone performance at wind speeds ranging up to 43.6 kilometres/hour.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5MTcyODkyLUNhbHRlY2hfTmV1cmFsX0ZseV8wOC5tYXgtMTQwMHg4MDAuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib"><h2 dir="ltr">Eating robot</h2>This news entered a new category. A food-tasting robot? Yes, a food-tasting robot. Researchers from the <a href="https://www.bbc.co.uk/news/uk-england-cambridgeshire-61313767" target="_blank">University of Cambridge&nbsp;</a>trained Beko to assess the saltiness of a dish at different stages of the chewing process, imitating a process that is similar in humans. Beko, who has already been trained to make omelettes, tasted nine variations of scrambled eggs and tomatoes at three different stages of the chewing process. The researchers attached a conductance probe to the robot&rsquo;s arm, which acts as a salinity sensor.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5MjAzNzIwLXJvYm90IGVhdGluZyBvbWVsZXR0ZXMuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib">Their results showed a significant improvement in robots&rsquo; ability to assess saltiness over other electronic tasting methods. In case you were wondering, the food was blended to imitate the change in texture caused by chewing.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5NTYyNTcwLVJvYm90LWNoZWYtdHJhaW5lZC10by10YXN0ZS1mb29kLWluLWRpZmZlcmVudC1jaGV3aW5nLXBoYXNlcy5qcGciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib"><h2 dir="ltr">Open source robotics &ndash; white papers</h2>We recently published a <a href="https://ubuntu.com/engage/robot-operating-system-choice" rel="noopener noreferrer" target="_blank">whitepaper about the importance of choosing the right OS</a> for your robot. Picking the right OS could make a big difference in factors like security, maintenance and infrastructure for deployment. Review the implications and explore why so many developers chose Ubuntu.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5NTc3MDcwLXdoaXRlcGFwZXIucG5nIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib"><h2 dir="ltr">Open source robotics &ndash; video news</h2>Another month, another video. More news, more applications, more robots! &nbsp;<iframe width="640" height="360" src="https://www.youtube.com/embed/hvmwE6ft45A?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe> If you want us to feature your robotics application, just send us an email at <a href="mailto:robotics.community@canonical.com" target="_blank">robotics.community@canonical.com</a>.<h2 dir="ltr">Open-source robotics &ndash; ROS 2 in Ubuntu with LXD tutorial</h2>Do you want to test ROS 2 Humble? If you want to test compatibility and keep your current Ubuntu (20.04, 18.04,&hellip;) environment stable until you know you are ready to upgrade, you can do that with LXD containers. <a href="https://ubuntu.com/blog/install-ros-2-humble-in-ubuntu-20-04-or-18-04-using-lxd-containers" target="_blank">Read our tutorial</a> to find out how.&nbsp;<h2 dir="ltr">Robot Makers - Chapter 2</h2>Did you miss the interview with Akara Robotics CEO Conor McGinn? Don&rsquo;t worry, the video is now available on-demand. If you want to learn about the funding landscape for robotics startups, the best sources of funding and good guidelines, this webinar is for you.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU2OTI5NjIxMzUzLVtXZWJpbmFyXSBBa2FyYSBSb2JvdCBNYWtlcnMgMiAoOCkucG5nIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib"><h2 dir="ltr">Stay tuned for more robotics news</h2>As always, we would love to hear from you. Send a summary of your robotics innovation and project to <a href="mailto:robotics.community@canonical.com" target="_blank">robotics.community@canonical.com</a>, and we will share it in our next robotics newsletter or monthly video. Thanks for reading!

Monthly robotics news blog. Discover the latest products, research, tutorials, and security news for robotics.

The State of Robotics, May & June 2022

Swarming robots hold the potential to solve a range of real-world problems, from <a href="https://www.wevolver.com/article/dlr-measures-flow-phenomena-around-wind-turbines-with-a-swarm-of-drones" target="_blank">analyzing airflow around wind turbines</a> and <a href="https://www.wevolver.com/article/adaptive-swarm-robotics-could-revolutionize-smart-agriculture" target="_blank">improving agricultural data gathering</a> to <a href="https://www.wevolver.com/article/nasa-works-to-give-satellite-swarms-a-hive-mind" target="_blank">monitoring the weather</a> or <a href="https://www.wevolver.com/article/could.tiny.satellites.explore.space.by.flying.in.flocks" target="_blank">exploring space</a>. There&rsquo;s only one problem: Research into robotic swarms can be an expensive endeavor.That&rsquo;s the problem HeRo 2.0 aims to solve. The creation of a quartet of roboticists from Brazil, HeRo 2.0 &mdash; an upgraded, streamlined version of the earlier HeRo platform from the same team &mdash; offers a low-cost compact robotics platform for swarm research and education based on off-the-shelf components and an easily-assembled 3D-printed chassis. Better still: The whole project is open source.<h1>Miniature marvels</h1>&ldquo;The robotic platform presented in this work takes advantage of the recent technological advancements to use mass-produced components that are smaller, affordable, and long-term available,&rdquo; the team explains of the robot. &ldquo;In addition, the design and assembly process follow new trends, such as the Maker Movement and Do It Yourself, which allow others to reproduce and customize the robot using additive manufacturing.&rdquo;The team set about designing HeRo 2.0 with a view to six key design principles: The robot should be as inexpensive as possible, to allow for larger swarms to be built without great expense; each individual robot should be small yet include on-board sensing capabilities and enough power for hours of autonomous operation; it should be highly tolerant to faults; communication between robots should be scalable to a large number of individuals; the robots should be easy to assemble and use only readily-available components; and they should be easy to program using existing development frameworks.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjQ5Mjc1NTczODU4LWhlcm8yLTIuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib" alt="A diagram of the HeRo 2.0 robotics platform, with the following parts labelled from the top: e-Hat, PCB board, cover, rotary encoder, battery, pinion gear encoder, rubber O-ring, servo motor SG90, castor wheel, motor chassis, board chassis, motor gear, spur gear wheel.">Built using off-the-shelf components in a 3D-printed chassis, HeRo 2.0 costs just $18 yet includes all the key functionality required for basic swarm research.The final design, which first author Paulo Rezeck and colleagues say meets all six of the requirements, is certainly low cost: The full bill of materials for each unit comes out at just $18.72, including $0.10 for a 38mm rubber O-ring and a battery powerful enough to push the robot for &nbsp;three hours of active experimentation &mdash; and that&#39;s a cost which could drop in large-volume production.The bargain-basement price comes courtesy of some unusual design choices, such as turning to a low-cost rotary encoder usually used to track the movement of the scroll wheel on a computer mouse to instead track the turning of the wheels. Continuous servo motors supporting simple control via the pulse-width modulation (PWM) output of a microcontroller further reduce costs, while the 3D-printable chassis uses just a little plastic and some electricity to be produced.<iframe src="https://www.youtube.com/embed/hAS7FKYkKWQ?&wmode=opaque&rel=0" allowfullscreen="" class="fr-draggable" width="640" height="360" frameborder="0"></iframe> The microcontroller, too, was picked for its high performance at a low price: An Espressif ESP-12E, based on the ESP8266 and boasting a single 32-bit Tensilica LX106 core running at speeds of up to 160MHz plus 4MB of flash memory. To this, the team added eight infrared sensors &mdash; officially limited to measuring distances no larger than 2cm (around 0.8&quot;), but extended through clever operation to 30cm (11.8&quot;). Further expansion, meanwhile, is possible through what the team calls &ldquo;e-Hats,&rdquo; add-on modules which attach to the top of the robot to provide additional functionality &mdash; such as, the team has proven in prototype, a display screen or an inertial measurement unit (IMU).<h1>Programming perfection</h1>On the software side, the HeRo 2.0 platform offers a choice of programming paradigm. Its firmware was developed on the Arduino integrated development environment (IDE) using the Wiring language, an extended variant of C designed for embedded work. The &ldquo;rosserial&rdquo; framework was added to provide compatibility with the Robot Operating System (ROS), a popular platform for robotics work.As a result, the robots can be programmed in two ways. For direct programming, a new firmware can be written in the Arduino IDE and flashed to the robots over a Wi-Fi connection using the Firmware Over the Air (FOTA) protocol, allowing algorithms to run directly on the robots&rsquo; microcontrollers. Otherwise, the stock firmware&rsquo;s ROS support can be leveraged to place the swarm under control of a centralized server communicating via ROS messages.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjQ5Mjc1NjM0MTQyLWhlcm8yLTQuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib" alt="A screenshot of the RViz software showing a map generated by a HeRo 2.0 robot; its path can be seen as a continuous line surrounded by data from its on-board distance sensors.">The researchers put the robot through a range of experimental tasks, including mapping, flocking, and collaboration on an object-transportation task.&ldquo;This mode is very convenient and scalable in the early stages of testing with multiple robots,&rdquo; the team notes of the server-based ROS programming mode. &ldquo;Furthermore, using the ROS framework for implementation, we have a series of tools and may use different programming languages.&rdquo;To prove the overall performance of the platform, the team set about comparing its capabilities to the E-puck &mdash; a highly popular commercial robot which costs $975 per unit, making it an order of magnitude more expensive than the HeRo 2.0 platform albeit with additional functionality including Bluetooth connectivity, additional sensors, and an on-board camera.In all tests the HeRo 2.0 proved capable of keeping up with its commercial rival, though with one caveat: While testing functionality like flocking and cooperative transportation, the team extended the robots&rsquo; on-board sensing capabilities with an overhead camera and the Apriltag tracking algorithm &mdash; to, as the team puts it, &ldquo;emulate a sensor in the robot that captures its neighbors&rsquo; position and relative velocity&rdquo; which is lacking in the hardware as-implemented.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjQ5Mjc1NjQ3MDQ4LWhlcm8yLTEuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 300px;" class="fr-fic fr-dib" alt="Four photographs of a HeRo 2.0 prototype shown from the top with its circuit board and distance sensors, from underneath with its wheels and castors visible, from the front, and from the side.">The compact HeRo 2.0 shows real promise for lowering the barrier to entry for swarm research, thanks to accessible development tools and its extremely low cost. There&rsquo;s room for improvement elsewhere, too. Proposed enhancements include a migration to the newer ROS 2 platform on the software side, the development of additional e-Hat add-ons in addition to the display and IMU units already prototyped, improvements to the internal filters for localization, and investigations in how to reduce variability of the 3D-printed parts in order to reduce post-processing and thus streamline assembly. More studies are also required to see how the low-cost parts stand up to repeated use &mdash; the rotary encoders in particular requiring replacement &ldquo;after months of use.&rdquo;The team&rsquo;s work has been submitted to the journal <cite>Autonomous Robots</cite>, with a preprint available <a href="https://arxiv.org/abs/2202.12391" target="_blank">on Cornell&rsquo;s arXiv server</a>.<h1>Reference</h1>Paulo Rezeck, Hector Azpurua, Mauricio FS Correa, and Luiz Chaimowicz: <cite>HeRo 2.0: A Low-Cost Robot for Swarm Robotics Research</cite>. DOI <a href="https://arxiv.org/abs/2202.12391" target="_blank">arXiv:2202.12391</a> <a href="http://cs.RO" target="_blank"></a>[cs.RO].

Costing just $18 each, the HeRo 2.0 robot platform includes expandability via "e-Hat" add-on hardware installed on the top.

Designed using off-the-shelf parts, a low-cost microcontroller, and a 3D-printed chassis, the HeRo 2.0 robot costs just $18 to build — and is at the heart of an effort to open swarm robotics up to a broader audience.

HeRo 2.0, an ultra-low cost 3D-printed robotics platform, could open swarm robotics experimentation up to all

At READY we want to engage the landscape of developers outside the robotics industry to be programmers in the robotics industry. &nbsp;People who focus on user experience design, on cloud platforms, on data infrastructure, on machine learning who are playing in the app space or the SaaS business space should all be encouraged to work in the industrial space. &nbsp;There is a huge potential of amazing things in the industrial robot and manufacturing space that they could contribute to and solve hard, interesting problems right now.&nbsp;So how do they find out how they can contribute? Well frankly, it’s something that, as an industry, we’re not doing the best at. If you look at just a some Google searches, and these are in quotes, you can see how well the Android and Windows spaces enable and provide starting guidance to developers:<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+ios+app+development%E2%80%9D" target="_blank">“Getting started with ios app development”</a> : 139,000<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+android+app+development%E2%80%9D" target="_blank">“Getting started with android app development”</a> : 262,000<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+window+program+development%E2%80%9D" target="_blank">“Getting started with window program development”</a> : 968,000,000<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+mac+program+development%E2%80%9D" target="_blank">“Getting started with mac program development”</a> : 697,000,000These results are not magic. If you search for other terms like “getting started with iOS development for an iPhone”. “Getting started with Android app development”, “getting started with windows program development”, you see millions, or at least hundreds of thousands, if not millions of results for resources. If a person asks, “Hey, I’m interested in building an app for an iPhone. Where do I start?”, and they find plenty of help.The robotics industry does not have this level of support, not even close. &nbsp;Just look at these search results (as of Sept 9, 2020)!“Getting started with robot app development” : 0“Getting started with robot application development” : 0“Getting started with robot program development” : 0We’re not telling a clear message about how people outside our industry can get started and create value in robot application development or robot program development. If we broaden the search out a little more, “getting started with robotics” we at least get a few hits, but those results are mostly in the hobbyist space.&nbsp;<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+robotics%E2%80%9D" target="_blank">“Getting started with robotics”</a> : 17,000<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+robot+programming%E2%80%9D" target="_blank">“Getting started with robot programming”</a> : 6<a href="https://www.google.com/search?q=%E2%80%9CGetting+started+with+industrial+robots%E2%80%9D" target="_blank">“Getting started with industrial robots”</a> : 1If I’m going to build a robot with an Arduino, then I can find lots of help. That’s awesome that people have found an accessible way to get into the industry through the hobbyist market. But when we talk about real industrial systems, there is next to no accessible content or support for the general developer. Whereas in the windows app development space, I can create almost any program with the documentation that’s out there. I can create a real enterprise grade application by myself. This is a story we’ve seen where people have created very impactful applications with very little resources and it made breakout success because they could create something that was industry grade with the tools that were available.&nbsp;How do we enable these developers? &nbsp;Well, a true developer ecosystem exists when there is a clear path for those in other industries to get excited about this industry. And that’s something that, as an outsider, isn’t apparent. As a developer I’d think, “I don’t know where to start”, “I don’t know what value it is”. They either look at manufacturing and they don’t see how high tech and interesting it has become. But if they get a chance, then they get in and they’re like, “Oh, this is so fragmented. I don’t know where to start” and that is too bad. It really comes down to focusing on platforms that reduce the onboarding costs of a developer to get into our ecosystem and start producing something that is amazing.You can build one piece of software that is widely deployed. Every startup I talked to in the robotic space, one of their major sort of efforts has been, “Oh, and I had to write my program so it can interface with this robot and this robot and this robot”. &nbsp;I hear this challenge all the time. &nbsp;There’s no path for them to build one program and it just works ubiquitously on all robots and peripherals as there is in the computer industry or the mobile industry. &nbsp;Quite simply, this challenge is a turnoff for developers because they don’t want to build the same thing repeatedly. They want to focus on what their value proposition is. They want to just build the thing they want to make and start providing value immediately.So what does a thriving developer ecosystem produce? Well, it produces useful robotic solutions that end users can use without direct help. &nbsp;It produces plugins and sub platforms that other developers can use. So that’s a key piece of the mobile and other software platform spaces is that there are developers who make plugins for other developers, and innovative hardware.At READY we are building an ecosystem where developers have very easy access to the tools that let them build industrial applications. They can start solving problems, they can start to monetize those solutions. And then the end users benefit by having these amazing applications that they can use to take their robot and really get stuff done with it, which in turn enables everyone else. It also enables the plugin developers, the hardware manufacturers, who are now more excited about the vibrancy of this ecosystem. It just raises everyone up.&nbsp;

At READY we want to engage the landscape of developers outside the robotics industry to be programmers in the robotics industry. 

How to Get Started in Robotic App Development

The automation of a production line has become a major issue for many industries, and the need to increase productivity while having a good return on investment has caused many concerns. The main area in which these challenges are tackled is bin-picking.Moving a part from a container to place it to a specific position, ready for the next step, sounds very simple, but it becomes complex when you need to abide to cycle-time requirements.As demand for accurate, autonomous and flexible bin-picking systems increase, Visio Nerf has produced a handbook to the technology designed introduce key concepts and offer concrete examples of cutting-edge technology. Below are some of thehighlights from the handbook. The book guide can be downloaded <a href="https://content.visionerf.com/from-2d-to-3d-the-operating-principles-of-industrial-vision-systems-0?hsCtaTracking=59e2a72a-d663-450a-af5e-51d259af5ed1%7C0aa52600-51c0-4eb1-a0a5-d30cb78dc3a7" rel="noopener noreferrer" target="_blank">here.&nbsp;</a><h2>Introduction to Bin-Picking</h2>The automation of production cells is at the heart of current concerns for many industries, the main objective being to find the right balance between investment and improved productivity. One of the operations for which the evolution towards a high-performance automated production tool has become necessary is Bin-Picking.&nbsp;Bin-picking involves moving a part from its original position to a specific delivery zone to make it available for the next production step. The workpiece is initially placed in a container or on a conveyor in a random, semi or fully-ordered manner depending on the application. The main challenge of Bin-Picking is to offer an efficient automated solution to meet cycle time optimization requirements, to protect operators from difficult and repetitive tasks particularly in harsh environments while increasing the flexibility and productivity of the work cell. Technological advances in the field of machine vision and calculation software can now provide solutions adapted to most industrial applications.<h2>Essential criteria for assessing bin-picking</h2>In order to assess the efficiency of a Bin-Picking cell and determine its dimensioning, several criteria must be taken into account:&nbsp;<ul><li>Cell cycle time&nbsp;</li><li>Bin emptying rate&nbsp;</li><li>Part pick and place accuracy</li></ul>These elements aim to measure and improve, more or less directly, the profitability of the cell at different times of its cycle life. A fourth factor to consider, that is too often overlooked, is the flexibility of the cell or its ability to be reused in several applications.With Industry 4.0, companies are looking for agile cells that can be reused without too many modifications when the type of the parts to be picked changes. These four criteria are intrinsically linked. For example, a suboptimal emptying rate may require the addition of mechanical systems to shake the bin or a change to more appropriate tools to complete the task. Integration of such elements extends the cycle time and penalizes the flexibility of the cell. Appropriate bin-picking cell sizing is essential and involves finding the right balance between these criteria. &nbsp;To read more about the technology behind bin-picking, download the full guide, written by Adib Rebbouh with Visio-Nerf&nbsp;<a href="https://content.visionerf.com/from-2d-to-3d-the-operating-principles-of-industrial-vision-systems-0?hsCtaTracking=59e2a72a-d663-450a-af5e-51d259af5ed1%7C0aa52600-51c0-4eb1-a0a5-d30cb78dc3a7" rel="noopener noreferrer" target="_blank">here</a>.&nbsp;

This handbook introduces the concept of bin-picking, and explores its complexity through real-case studies.

The Bin-Picking Handbook For Engineers

Robots are increasingly becoming a part of our society and although right now they&rsquo;re just predominantly deployed in industrial warehouses and automation lines in factories, robots with additional safety features and environment-aware motion planning are at the very cutting edge of robotics research. These collaborative robots or cobots for short, are specially designed to work alongside humans in an unrestricted environment where humans can freely move around.&nbsp;These cobots when combined with Artificial Intelligence or AI, can effectively work alongside humans in the real world and help humans with their day-to-day tasks. AI helps computers make smart decisions to solve problems efficiently. &nbsp; &nbsp;Usually, warehouses are what roboticists call a structured environment, meaning, everything is exactly where it was always placed, and the task is repetitive and has no dynamically changing parameters. For such a task, the involvement of AI would be overkill and can easily be programmed from start to end using simple code. A semi-structured environment is something like a hospital, where most things have a fixed place, there are regulations set about the movement of people and goods around the area, however, there are certain aspects that are volatile and non-repetitive that need dynamic decision making. An example of an unstructured environment is a vegetable sorting line conveyor system, where the throughput of the vegetables is variable, the number of good and bad items varies with time and their locations on the conveyor are dynamically changing all the time.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjIxNjQ1NTMzNjA2LTEyLmpwZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 716px;" class="fr-fic fr-dib">This is one of the hardest problems to solve as it involves so many different aspects of robotics, such as AI for deciding when to do which action, path planning, and trajectory optimization, for finding ways to perform that action, control systems, for accurately following that plan and computer vision, to identify, track and provide 3D coordinates for the robot to reach. This is the problem THINC Lab researchers are currently engaged in solving.&nbsp;To give a brief overview of the complete architecture, we have an RGBD camera observing the items on a conveyor belt, the RGB or color images are fed into a trained deep learning neural network that identifies the salient objects in the image. These object locations are then identified in the real world using computer vision transformation techniques. In the following image, you can see the cobot Sawyer, a conveyor belt with onions, and a bin to place the bad onions recognized by the neural network. From here, some robot programming as mentioned before will do the job of sorting the onions.<iframe src="https://www.youtube.com/embed/6yAhe-vZvAk?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" allowfullscreen="" class="fr-draggable" width="640" height="360" frameborder="0" data-gtm-yt-inspected-8="true" data-lf-form-tracking-inspected-kn9eq4roawkarlvp="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" data-lf-vimeo-playback-inspected-kn9eq4roawkarlvp="true"></iframe> However, how does the robot know what to do after identifying and picking up an object? The task may sometimes be too complex to explicitly program. This is where AI steps in again. Using an AI technique called Inverse Reinforcement learning (IRL), the robot can observe a human performing the task and learn how to do it by itself.Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/how-far-along-are-we-in-the-ai-human-race" rel="noopener noreferrer" target="_blank">How far along are we in the AI-Human &quot;race&quot;?</a>The idea behind IRL is to learn the intent of the expert performing the task. Once that is understood, the robot can do it in a way that it deems plausible while preserving the intent. The following image shows another deep-learning network called the SA-Net processing the data of the human performing the task and extracting the salient features needed to learn.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjIxOTU3MDYyMjU2LXNhbmV0LnBuZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 660px;" class="fr-fic fr-dib">The system is still in its prototype stage and this proof of concept setup would soon be improved to the stage of commercial deployment. &nbsp;For further details about the research we do, check out the <a href="http://thinc.cs.uga.edu/" rel="noopener noreferrer" target="_blank">THINC Lab website</a>.

Cobots when combined with Artificial Intelligence(AI), can effectively work alongside humans in the real world on their daily tasks. With an AI technique called Inverse Reinforcement learning (IRL), the robot can observe a human performing the task and learn how to do it quite intuitively.

Can Robots watch Humans  and learn a task?

What is No Code robotic programming? Kel Guerin, PhD roboticist and CTO and Co-founder of READY Robotics, will explain what No Code programming is, what makes it different from traditional robot programming, and how it can make robots and automation more accessible to all manufacturers. &nbsp;Learn more about No Code programming: Download the whitepaper to learn more about No Code programming and how it is democratizing automation.&nbsp;<a href="https://www.ready-robotics.com/resources/whitepapers/the-no-code-revolution" rel="noopener noreferrer" target="_blank">https://www.ready-robotics.com/resources/whitepapers/the-no-code-revolution</a>

What is No Code robotic programming? Kel Guerin, PhD roboticist and CTO and Co-founder of READY Robotics, will explain what No Code programming is.

What Does Robotic No-Code Programming Look Like

For outdoor robotics, one of the key capabilities for autonomous use-cases is GPS Waypoint Navigation.If you&rsquo;re unfamiliar with the term, GPS Waypoint Navigation is the ability to provide a robot with a set of GPS waypoints (i.e., a set of latitude / longitude pairs), and have the robot autonomously navigate from its current location to each of the defined waypoints. An added bonus would be not bumping into anything along the way. This has been done in research projects and papers a number of times, but scalable hardware and software packages to enable this on production ready field robots are few and far between.That&rsquo;s why Clearpath and <a href="https://veerum.com/" target="_blank">Veerum,&nbsp;</a>with the help of a <a href="http://mitacs.ca/en" target="_blank">Mitacs</a> grant, partnered with Steve Phillips, a civil engineering graduate student from the University of Waterloo. Steve&rsquo;s research is a cross between mechanical and structural engineering, focusing on using mobile robots to perform infrastructure assessment and inspection. For this project with Clearpath, he&rsquo;s developed a robust, repeatable GPS Waypoint Navigation system for Clearpath&rsquo;s family of outdoor robots.Over the past semester, Steve has been working away at selecting and integrating the right payload suite, perfecting the fusion of sensor data, and creating a usable interface to easily select a waypoint and send the robot on its way.<h3>THE SETUP</h3>For development and testing of the GPS navigation software, Steve&rsquo;s primary setup consisted of a <a href="https://clearpathrobotics.com/jackal-small-unmanned-ground-vehicle/" target="_blank">Jackal UGV</a>, paired with the rugged <a href="https://www.swiftnav.com/duro" target="_blank">Duro GPS from Swift Navigation</a> and the UM7 inertial measurement unit (IMU) from Redshift Labs. In addition to supporting high accuracy RTK, the Duro includes a built in Bosch <a href="https://www.bosch-sensortec.com/bst/products/all_products/bmi160" target="_blank">BMI160 IMU&nbsp;</a>and <a href="https://www.bosch-sensortec.com/bst/products/all_products/bmm150" target="_blank">BMM150 Magnetometer</a>. To verify that the software functions hardware agnostically, Steve and the Clearpath software team also deployed it on the much larger <a href="https://clearpathrobotics.com/warthog-unmanned-ground-vehicle-robot/" target="_blank">Warthog UGV</a> (outfitted with a Novatel GPS and a Lord GX5-25 IMU). Running autonomy software on a 500 lb robot can be scary, so we&rsquo;re glad that the system was thoroughly tested beforehand, and works perfectly!GPS waypoint navigation consists of two key components: localization, and navigation. Localization is responsible for using the robot sensors (i.e., the UGV wheel encoders, IMU data, and gps measurements) to estimate the position of the robot. Navigation is responsible for sending the UGV wheel velocity commands which move it to the desired destination (e.g., a waypoint). By utilizing a combination of tried-and-true ROS packages, as well as some new &amp; cutting edge navigation packages unveiled at ROSCON 2017 (see <a href="https://vimeo.com/236174072" target="_blank">this video</a> and <a href="https://vimeo.com/236487972" target="_blank">this video</a> for more information), the GPS waypoint navigation framework is robust and reliable, but also highly configurable and easily extensible to incorporate updates.<h3>MISSIONS</h3>Another important component of the GPS navigation solution is the interface used to define and send waypoints, as well as visualize the robot and its environment as it navigates. In the current framework, users create (or load existing) &ldquo;missions&rdquo;, which are sent to the robot to initiate navigation.A mission is defined as follows:<ul><li>Every mission is composed of a set of waypoints</li><li>Each waypoint is defined by:<ul><li>A position (latitude, longitude, heading)</li><li>A set of &lsquo;viapoints&rsquo;, which define the path that the robot should take to arrive at the waypoint. Viapoints are used to ensure that the robot takes the desired route between waypoints, avoiding any non-moving obstacles (like buildings) or other off-limit areas.</li><li>A list of tasks to execute once the robot arrives at the waypoint (e.g., take a photo, collect a 3d scan, wait, etc.)</li></ul></li></ul>With this framework, the concept of simple point-to-point navigation has been extended to provide a solution that better meets the requirements we see in real world applications.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1579090342310-GPS-Waypoint.gif">By using the familiar, top-down, satellite view we see in Google Maps, setting up autonomy missions for the robot is very intuitive.&nbsp;<h3>APPLICATION SOFTWARE PACKAGE SUITE</h3>While Steve was only with Clearpath for the semester, his work has become part of Clearpath&rsquo;s line of application&nbsp;<a href="https://clearpathrobotics.com/gps-waypoint-navigation-package/" target="_blank">Software Packages</a>. These packages are designed to enable autonomy and teleoperation across a number of different applications, and the GPS Waypoint Navigation stack perfectly complements the other available packages. &nbsp;If you&rsquo;re working on an application that could benefit from outdoor autonomy, take a look at the GPS Waypoint Navigation package, as well as the other available autonomy software.

If you’re unfamiliar with the term, GPS Waypoint Navigation is the ability to provide a robot with a set of GPS waypoints (i.e., a set of latitude / longitude pairs), and have the robot autonomously navigate from its current location to each of the defined waypoints.

A Better Solution For GPS Waypoint Navigation

We are announcing the release of our state-of-the-art off-policy model-free reinforcement learning algorithm, soft actor-critic (SAC). This algorithm has been developed jointly at UC Berkeley and Google, and we have been using it internally for our robotics experiment. Soft actor-critic is, to our knowledge, one of the most efficient model-free algorithms available today, making it especially well-suited for real-world robotic learning. In this post, we will benchmark SAC against state-of-the-art model-free RL algorithms and showcase a spectrum of real-world robot examples, ranging from manipulation to locomotion. We also release our implementation of SAC, which is particularly designed for real-world robotic systems.<h1>Desired Features for Deep RL for Real Robots</h1>What makes an ideal deep RL algorithm for real-world systems? Real-world experimentation brings additional challenges, such as constant interruptions in the data stream, requirement for a low-latency inference and smooth exploration to avoid mechanical wear and tear on the robot, which set additional requirement for both the algorithm and also the implementation of the algorithm.Regarding the algorithm, several properties are desirable:<ul><li>Sample Efficiency. Learning skills in the real world can take a substantial amount of time. Prototyping a new task takes several trials, and the total time required to learn a new skill quickly adds up. Thus good sample complexity is the first prerequisite for successful skill acquisition.</li><li>No Sensitive Hyperparameters. In the real world, we want to avoid parameter tuning for the obvious reason. Maximum entropy RL provides a robust framework that minimizes the need for hyperparameter tuning.</li><li>Off-Policy Learning. An algorithm is off-policy if we can reuse data collected for another task. In a typical scenario, we need to adjust parameters and shape the reward function when prototyping a new task, and use of an off-policy algorithm allows reusing the already collected data.</li></ul>Soft actor-critic (SAC), described below, is an off-policy model-free deep RL algorithm that is well aligned with these requirements. In particular, we show that it is sample efficient enough to solve real-world robot tasks in only a handful of hours, robust to hyperparameters and works on a variety of simulated environments with a single set of hyperparameters.In addition to the desired algorithmic properties, experimentation in the real-world sets additional requirements for the implementation. Our release supports many of these features that we have found crucial when learning with real robots, perhaps the most importantly:<ul><li>Asynchronous Sampling. Inference needs to be fast to minimize delay in the control loop, and we typically want to keep training during the environment resets too. Therefore, data sampling and training should run in independent threads or processes.</li><li>Stop / Resume Training. When working with real hardware, whatever can go wrong, will go wrong. We should expect constant interruptions in the data stream.</li><li>Action smoothing. Typical Gaussian exploration makes the actuators jitter at high frequency, potentially damaging the hardware. Thus temporally correlating the exploration is important.</li></ul><h1>Soft Actor-Critic</h1>Soft actor-critic is based on the maximum entropy reinforcement learning framework, which considers the entropy augmented objectiveJ(&pi;)=E&pi;[&sum;tr(st,at)&minus;&alpha;log(&pi;(at|st))],where&nbsp;st and&nbsp;at are the state and the action, and the expectation is taken over the policy and the true dynamics of the system. In other words, the optimal policy not only maximizes the expected return (first summand) but also the expected entropy of itself (second summand). The trade-off between the two is controlled by the non-negative temperature parameter&nbsp;&alpha;, and we can always recover the conventional, maximum expected return objective by setting&nbsp;&alpha;=0. In a&nbsp;<a href="https://arxiv.org/abs/1812.05905" target="_blank">technical report</a>, we show that we can view this objective as an entropy constrained maximization of the expected return, and learn the temperature parameter automatically instead of treating it as a hyperparameter.This objective can be interpreted in several ways. We can view the entropy term as an uninformative (uniform) prior over the policy, but we can also view it as a regularizer or as an attempt to trade off between exploration (maximize entropy) and exploitation (maximize return). In our&nbsp;<a href="https://bair.berkeley.edu/blog/2017/10/06/soft-q-learning/" target="_blank">previous post</a>, we gave a broader overview and proposed applications that are unique to maximum entropy RL, and a probabilistic view of the objective is discussed in a&nbsp;<a href="https://arxiv.org/abs/1805.00909" target="_blank">recent tutorial</a>. Soft actor-critic maximizes this objective by parameterizing a Gaussian policy and a Q-function with a neural network, and optimizing them using approximate dynamic programming. We defer further details of soft actor-critic to the&nbsp;<a href="https://arxiv.org/abs/1812.05905" target="_blank">technical report</a>. In this post, we will view the objective as a grounded way to derive better reinforcement learning algorithms that perform consistently and are sample efficient enough to be applicable to real-world robotic applications, and&mdash;perhaps surprisingly&mdash;can yield state-of-the-art performance under the conventional, maximum expected return objective (without entropy regularization) in simulated benchmarks.<h1>Simulated Benchmarks</h1>Before we jump into real-world experiments, we compare SAC on standard benchmark tasks to other popular deep RL algorithms, deep deterministic policy gradient (DDPG), twin delayed deep deterministic policy gradient (TD3), and proximal policy optimization (PPO). The figures below compare the algorithms on three challenging locomotion tasks, HalfCheetah, Ant, and Humanoid, from OpenAI Gym. The solid lines depict the total average return and the shadings correspond to the best and the worst trial over five random seeds. Indeed, soft actor-critic, which is shown in blue, achieves the best performance, and&mdash;what&rsquo;s even more important for real-world applications&mdash;it performs well also in the worst case. We have included more benchmark results in the <a href="https://arxiv.org/abs/1812.05905" target="_blank">technical report</a>.<h1>Deep RL in the Real World</h1>We tested soft actor-critic in the real world by solving three tasks from scratch without relying on simulation or demonstrations. Our first real-world task involves the Minitaur robot, a small-scale quadruped with eight direct-drive actuators. The action space consists of the swing angle and the extension of each leg, which are then mapped to desired motor positions and tracked with a PD controller. The observations include the motor angles as well as roll and pitch angles and angular velocities of the base. This learning task presents substantial challenges for real-world reinforcement learning. The robot is underactuated, and must therefore delicately balance contact forces on the legs to make forward progress. An untrained policy can lose balance and fall, and too many falls will eventually damage the robot, making sample-efficient learning essentially. The video below illustrates the learned skill. Although we trained our policy only on flat terrain, we then tested it on varied terrains and obstacles. Because soft actor-critic learns robust policies, due to entropy maximization at training time, the policy can readily generalize to these perturbations without any additional learning.<iframe src="https://www.youtube.com/embed/KOObeIjzXTY?&wmode=opaque" frameborder="0" allowfullscreen="" class="fr-draggable" style="width: 1114px; height: 631px;"></iframe>The Minitaur robot (Google, Tuomas Haarnoja, Sehoon Ha, Jie Tan, and Sergey Levine).&nbsp;Our second real-world robotic task involves training a 3-finger dexterous robotic hand to manipulate an object. The hand is based on the Dynamixel Claw hand, discussed in <a href="https://bair.berkeley.edu/blog/2018/08/31/dexterous-manip/" target="_blank">another post</a>. This hand has 9 DoFs, each controlled by a Dynamixel servo-motor. The policy controls the hand by sending target joint angle positions for the on-board PID controller. The manipulation task requires the hand to rotate a ``valve&rsquo;&lsquo;-like object as shown in the animation below. In order to perceive the valve, the robot must use raw RGB images shown in the inset at the bottom right. The robot must rotate the valve so that the colored peg faces the right (see video below). The initial position of the valve is reset uniformly at random for each episode, forcing the policy to learn to use the raw RGB images to perceive the current valve orientation. A small motor is attached to the valve to automate resets and to provide the ground truth position for the determination of the reward function. The position of this motor is not provided to the policy. This task is exceptionally challenging due to both the perception challenges and the need to control a hand with 9 degrees of freedom.&nbsp;<iframe src="https://www.youtube.com/embed/ThlkphNtaWM??list=PLu1urizo_exRQcDIzUU-0wiWiMUDTVfFD&wmode=opaque" frameborder="0" allowfullscreen="" class="fr-draggable" style="width: 1115px; height: 730px;"></iframe>Rotating a valve with a dexterous hand, learned directly from raw pixels (UC Berkeley, Kristian Hartikainen, Vikash Kumar, Henry Zhu, Abhishek Gupta, Tuomas Haarnoja, and Sergey Levine).&nbsp;In the final task, we trained a 7-DoF Sawyer robot to stack Lego blocks. The policy receives the joint positions and velocities, as well as end-effector force as an input and outputs torque commands to each of the seven joints. The biggest challenge is to accurately align the studs before exerting a downward force to overcome the friction between them.&nbsp;Soft actor-critic solves all of these tasks quickly: the Minitaur locomotion and the block-stacking tasks both take 2 hours, and the valve-turning task from image observations takes 20 hours. We also learned a policy for the valve-turning task without images by providing the actual valve position as an observation to the policy. Soft actor-critic can learn this easier version of the valve task in 3 hours. For comparison,&nbsp;<a href="https://arxiv.org/abs/1810.06045" target="_blank">prior work</a> has used PPO to learn the same task without images in 7.4 hours.<h1>Conclusion</h1>Soft actor-critic is a step towards feasible deep RL with real-world robots. Work still needs to be done to scale up these methods to more challenging tasks, but we believe we are getting closer to the critical point where deep RL can become a practical solution for robotic tasks. Meanwhile, you can connect your robot to our toolbox and get learning started!

The Minitaur robot (Google, Tuomas Haarnoja, Sehoon Ha, Jie Tan, and Sergey Levine).

We are announcing the release of our state-of-the-art off-policy model-free reinforcement learning algorithm, soft actor-critic (SAC). This algorithm has been developed jointly at UC Berkeley and Google, and we have been using it internally for our robotics experiment.