AI: changing the landscape of the robotics industry
In the past, the phrase “cutting-edge” may have implied teleoperated or semi-autonomous robots. Today, artificial intelligence, including computer vision and machine learning, are changing the landscape of the robotics industry. Tech analysts are deeming the blending of the physical and digital worlds the 4th Industrial Revolution— one that is sure to transform business operations across numerous industries.
To stay competitive in a global market, businesses are now considering robots that are fully autonomous—ones that can perceive, interact and conceptualize the world around them. As industries begin to navigate this modern technological revolution, they are searching for trusted and experienced technology partners. Our capabilities have expanded significantly beyond teleoperated robots and into the field of computer vision and machine learning. In other words, our latest generation of robots have the capacity to see, feel, and interact with the world, making them more human-like than ever before.
In this white paper, we will define and expound upon some of the common terminology being used in the field of autonomous robotics. We will discuss what is meant by the terms AI, computer vision, and machine learning and how these terms are being applied to modern robotics. In addition, we will identify some of the facts that are prompting businesses to pursue the use of AI in their operations. Finally, we will explain how RE2 Robotics is using AI, CV and ML to expand the capabilities of its next-generation of robotic arms.
First coined in the 1950s, the term “artificial intelligence” or “AI” sometimes has a negative connotation. In today's popular culture, the term is often associated with robots “taking over” by becoming more intelligent and sentient than the humans who built and programmed them. In reality, nothing could be further from the truth. Rather, the term “artificial intelligence” applies to an entire field of knowledge in which robotic systems are programmed to learn from and react to the world around them.
The term “artificial intelligence” incorporates the areas of computer science that involve machine learning, deep learning, and computer vision. These programming techniques give robots the ability to understand, learn, and perceive in a way that mimics the human brain. In essence, AI programming trains machines to perform human tasks.
Similar to how biology and chemistry fall under the general umbrella of science, artificial intelligence is an entire knowledge field that is comprised of separate, distinct areas of computer science. One of those areas is machine learning, a field of study that combines the principles of computer science and statistics to create statistical models and train machines how to learn. These models are generally used for two purposes: to make predictions about the future based on data about the past, and to make inferences to discover data patterns.
Teaching a machine to learn typically involves three primary components: data, features, and algorithms.
Machine learning requires large amounts of diverse data to find patterns and “learn.” There are two main ways to collect that data: manually or automatically. Manually collected data contains far fewer errors but takes more time to collect and therefore, is more costly, as it labor-intensive. The automatic approach, on the other hand, is less costly to conduct, since the system is gathering everything it can on its own and is essentially “hoping for the best.” Either way, the more diverse the data, the better the results will be.
Also known as parameters or variables, features are the building blocks of data. They provide the algorithms with the characteristics or properties needed to perform a given calculation. These characteristics are usually numerical, but they can also be graphical. Selecting the features that most distinctly characterize or represent the data is vital and often time consuming. The quality of the features selected has a major impact on the quality of the insights gained through machine learning. This process is difficult for a human to do by hand. In addition, the main source of error often results from these features not being describable or understandable by humans.
Feature-selection processes, or feature engineering, can improve the quality of one’s data. This is often a difficult process, but if performed well, the resulting data will contain the features essential to solving the problems. This, in turn, can lead to building the possible data models.
Machine learning relies upon different types of algorithms to learn. These include supervised methods such as classification and regression, as well as unsupervised methods like clustering or pattern search. First, we will examine supervised algorithms.
There are two main types of supervised machine learning:
With supervised algorithms, the machine has a “supervisor” or “teacher” who gives the machine all of the answers by labeling—or attributing an output—to each image by hand. For example, supervised learning allows the machine to separate cats from dogs, or pieces of luggage from a refrigerator, in a given photo. Supervised learning methods are often seen as complex in comparison to unsupervised methods, because the supervisor has to understand and label the inputs very precisely for success. This can be a time-consuming process.
Like most humans, machines typically learn faster with a teacher. Therefore, supervised machine learning is more commonly used with real-life tasks. With supervised machine learning, the data is pre-categorized or numerical.
In an unsupervised algorithm, on the other hand, there is no prior knowledge. The data is not known and is not labeled. As a result, the machine is left on its own to differentiate objects, which, in turn, can mean a decrease in accuracy of results when compared to supervised methods. Since unsupervised learning can solve complex problems using only input data, it gives machines the ability to truly learn on their own.
Clustering and pattern search are two commonly used methods of unsupervised machine learning. Clustering works by dividing objects into categories based on similarities and differences. The machine chooses the best way to label the data and divides objects into groups that share similar traits.
As the name implies, pattern search, or pattern recognition, is a method that looks for patterns and similarities in data. It enables a machine to categorize the data by matching the inputs to data already stored in a database.
Choosing the best algorithm for your data and features is critical. After all, the selected method affects the precision, performance, and size of the final model. If the data is bad or “dirty” (e.g., if fields are missing, or if it the data is duplicated or outdated), even the best algorithm will not generate accurate results.
Artificial neural networks are a method of machine learning that is used for both classification and regression. Modeled loosely on the human brain, neural nets form the building blocks of deep learning. In the last decade, neural nets have become a more widely used machine learning algorithm because they can produce state-of-the-art results in both image and speech recognition.
Neural nets operate by creating connections between “neurons,” which include the algorithm’s input values, weights and bias, sum, and activation functions.
Deep learning is a method of machine learning that relies upon a family of algorithms. Deep learning models use artificial neural nets to process large quantities of data— including images, text and sounds—to produce startlingly accurate results. In short, deep learning allows machines to learn by example in the same way that humans do.
Why modern industry is using AI
Why would modern businesses want to implement artificial intelligence into their operations? While many people think that AI and machine learning are limited to self-driving cars and voice-activated assistants like Amazon’s Alexa™ or Apple’s Siri™, the reality is that AI-driven automation has proven to be useful in a variety of applications across numerous industries, including the aviation, medical, agriculture, energy, and material handling markets, to name just a few.
For example, on the factory floor, AI is being used not only to automate tasks, but also to diagnose equipment malfunctions or detect product anomalies. Within the aviation industry, for example, AI and machine learning can be used to predict peak travel times, assist with passenger check-in, and automate routine maintenance tasks. Material handling operations can utilize AI to delineate and separate packages by size, shape, and color, minimizing processing errors and streamlining operations. Within the medical industry, AI is being used everywhere from patient registration to the operating room, where AI-enabled robots are assisting with surgery, leading to better accuracy and a faster recovery time.
Overall, AI is being used across the board to automate dangerous tasks, augment or replace skilled labor, and streamline operations. And yet, despite these advances, AI still does not have the ability to think abstractly or creatively—meaning that humans still need to be very much in the loop—and this creates a need for an ever-evolving workforce that is trained in robotics and advanced manufacturing.
Next generation AI capabilities
As more companies turn to robotics, AI, and machine learning to automate their processes, keeping a competitive edge in today’s ever-evolving global market is essential—and working with a company that has a proven track record of bringing next-generation technology to market is key.
Our newest robotic systems use third-party multi-modal 2D and 3D imaging sensors with RE2 proprietary algorithms that can perceive the world in both indoor and outdoor environments. Most perception systems that control robotic arms must operate in constrained indoor environments with controlled illumination. RE2’s perception systems, on the other hand, can detect and track objects in just about any indoor or outdoor environment.
In addition, RE2’s control algorithms “close the loop” by continually adjusting the robotic arm’s position and orientation based on real-time vision processing and reporting. These methods are accelerated by customizing the input and processing hardware and, unlike typical industrial solutions that rely on 3D models, our system leverages the geometric structure of objects and logic to enable unique grasps for varied scenarios.
RE2 is at the forefront of designing truly intelligent robotic systems. Using data collected from RE2 Detect systems, our autonomy algorithms fuse geometric computer vision with traditional machine learning and deep learning techniques to provide human-like decision processing.
Unlike traditional autonomy algorithms, which are based on a single method that can only work in structured environments with controlled lighting, RE2’s robust software can handle anomalies typically experienced in unstructured, outdoor environments with variable lighting—similar to the way human beings perceive and process information. In other words, our algorithms can learn and adapt in any environment, allowing users to harness the power of intelligent mobile manipulation.
Pushing the limit of what's possible
As robotic technology evolves, RE2 Robotics will continue to push the limits of what is possible with artificial intelligence. RE2 is uniquely poised to build upon its history of designing truly human-like robotic arms, both teleoperated and autonomous, to advance its proven capabilities even further. Whether you are considering robotics to disrupt your industry, supplement your workforce, or improve worker safety, the potential applications of artificial intelligence are endless.
As with any major organizational change, the implementation of AI requires a thorough evaluation of your business operations. This requires careful consideration of the challenges you would like to solve using AI. The successful implementation of machine learning and computer vision algorithms requires hefty amounts of data. Therefore, you will need to determine whether that data can be automatically or manually collected, whether you will use a supervised or unsupervised algorithm, and whether the algorithm you need already exists or if you will need a fusion of algorithms to solve your problem. in addition, we can help you determine which sensors would be the most appropriate for your specific needs.
Already known for its compact, dexterous, and strong robotic arms, RE2’s technology is using machine learning and computer vision to perceive the world in a variety of complex environments, both indoor and outdoor. These advancements are allowing businesses to automate their operations more than ever before. Today, across every industry, the question is no longer “how” a business will implement next-generation robotics, but “when.”