|Degrees of freedom (DOF)||12|
|- Lower body||40.24||kg|
|- Upper body||30.05||kg|
|Low reduction ratio||50:1|
|50:1||Intel I7-630m processor and 4Gbyte DDR3 RAM|
|Sensors (above the foot and below the ankle motor)||2||ATI DAQ MINI85 FT|
|Sensors (upper housing )||1||IMU 3DM-GX3-25|
The robot has human-like and compliant motion in multiple contact situations through the use of torque controlled joints. The aesthetic design is achieved by the robot design process using an integrated design and frame through multi-axis CNC machining.
A larger distance between the two legs was used in order to avoid collisions between the legs as it is one of the most effective and easiest ways to overcome these collisions. There are two legs with 6 degrees of freedom each, 3 axis of rotation in the hip joint, 1 in the knee, and 2 in the ankle.
This research presents the design process and performance metrics of the design in terms of weight, strength, range of motion, visual cues, and accessibility. The experimental results demonstrate the performance of torque-controlled joints during gravity compensation.
In the field of design, it is known that by just changing the aesthetic of an object, the perceived usability will change. Paolo Dario suggests in the field of personal robotics that the performance of a robot can be evaluated in the same way as a home appliance. As industrial robots have been designed largely for manufacturing facilities and with minimum thought of integration with humans, the knowledge gained in these areas about design cannot be directly transferred.
In addition, it is important to note that there may not be one specific design that succeeds above all in every category. The overall quality of aesthetic design is more important than the closeness to anthropomorphic appearance. Similarly, a difference in the acceptance of a robots’ design based on the task the robot is performing has been found.
Figure 2: Diagram of the link lengths used in the robot based off of an average Korean female. The lengths are adjusted slightly for simplicity in design and control. The joint order is listed on the left.
Advancements made in manufacturing, computer modeling, material properties, and electronics over the past 40 years have enabled researchers to develop more compact and able robots, such as the case of Honda P2 to P3. However, while current CAD/CAM techniques allow for more complex three-dimensional parts, very few humanoids have taken advantage of machining techniques beyond the third axis. Many research robots use two-dimensional parts to assemble a three-dimensional structure or have used a minimum amount of three-dimensional machining and put casings on the robot for a desired aesthetic. However, even with casings the robots rarely have an aesthetic design that moves beyond rectangular shapes.
Full scale human proportions were considered alongside the motor selection. Through a CAD/RP/CAM process the aesthetic design of the robot was achieved through the frame, eliminating the need for casings. Multi-axis machining was used in order to reduce structural components and create a unique and minimalistic design. The closeness to an anthropomorphic figure is not necessarily the gold standard and as such DYROS Humanoid was not designed to replicate a human but to reference the proportions. The aesthetics are important to the goal and was approached by an interdisciplinary team of industrial design and engineering.
The robot is designed as a torque-controlled robot with 12 DOF. A low reduction ratio of 50:1 is used for the motors which are directly connected to the joints. This gear ratio and direct connection are to provide minimum friction, as well as providing good back-drivability for compliant motion control without using joint torque sensors. A small motor size is chosen in order to maintain a thin leg. While the current robot consists only of legs, the technical specifications were chosen to include a full size upper body as well. The joint connections are, in order: Hip yaw, Hip roll, Hip pitch, Knee pitch, Ankle pitch, Ankle roll.
As the robot was developed in South Korea, the average Korean female proportions were taken as the baseline for creating the full scale legs. Link lengths and joint order locations are seen in Fig. 2. A larger distance between the two legs was used in order to avoid collisions between the legs as it is one of the most effective and easiest ways to overcome these collisions. There are two legs with 6 degrees of freedom each, 3 axis of rotation in the hip joint, 1 in the knee, and 2 in the ankle.
As a starting point for motor selection we based our simulation robot on MAHRU. The robot was simulated in the physics based simulation software RoboticsLab. Contact consistent whole-body control framework was used for the robot control in the simulation. The simulated robot weight was 71.295 kg: 40.245kg for the lower body and 30.05kg for the upper body. Both squat motion and walking motion were simulated. The squat motion was controlled up to a 141 degree bend of the knee joint. Squatting time was simulated at 1 second. Table I shows the results of the squat simulation.
The forward walking motion was controlled with a speed up to 0.3m/sec. In the walking simulation the COM position, foot position and orientation, and trunk orientation were controlled in the task space. The double support time was 0.3sec, single support time was 0.7sec, and the stride length was 0.1, 0.2, and 0.3 meters, of which the largest RMS value in all scenarios was taken. Simulation results of the walking motion are seen in Table II.
The motors are chosen as the ones in Table III such that the peak torque in squat and walking simulation is approximately two times the continuous torque and less than the peak torque of each motor. All joints have a Kollmorgen motor and harmonic gear. The upper body consists of a computer with an Intel I7- 630m processor and 4Gbyte DDR3 RAM. The computer has additional safety features to endure vibration while also being compact. The computer runs roboticsLab as a realtime control software. An AS5145 absolute encoder is connected to the joint link and a RMB incremental encoder is connected to the motor. In order to control each motor at the same time, the Elmo gold solo whistle digital servo drive was selected as the motor driver. EtherCAT is used for fast communication between the motor drive and computer. The robot has two ATI DAQ MINI85 FT sensors located above the foot and below the ankle motor. An IMU 3DM-GX3-25 is located in the upper housing with the computer.
After simulation to find the required motors, a simplistic box model was created with the proper placements of the motors. A common method for humanoid design is in maintaining this relatively box-like shape and using lightweight plastic casings to create the desired aesthetic. DYROS Humanoid was designed to integrate the frame and the design. The exclusion of coverings and an open frame design allow for more airflow in cooling the electronics. A combination of curved cylinders and plates are used to create a unique aesthetic while informing the observer of the intended human-like movement. Structural front plates act as a design component as well as a secondary heatsink. Through time invested in the multi-dimensional parts, both the aesthetic and structural components can be unified. Additionally, the assembly time, complexity of the robot, and maintenance difficulty are reduced.
Figure 4: Left: The first design iteration with two rotations showing an intersection of the link connections. Right: The revised design with a larger range of motion.
The links were given constraints in SolidWorks to view the range of motion. While the initial design of the structure provided the desired range of motion on each individual axis, a rotation on more than one axis at the same time showed interference by the link connections. The range of motion was extended by modifying some connections to the motors (Fig. 4). In line with the stress analysis, some parts were modified for their aesthetics as well as structure. The cylinders on the upper link were thickened from 18mm to 22mm to provide a stronger and more balanced visual weight to the solid front plate. The upper link plate was then reduced from 10.71mm to 7.99mm thick. The lower link cylinders remained almost the same, while the front plate was also reduced. Table IV shows the second iteration was stronger, more aesthetically consistent, and decreased in weight.
Describes the design process and performance metrics of the design in terms of weight, strength, range of motion, visual cues, and accessibility. Experimental results demonstrate the performance of torque-controlled joints during gravity compensation.