Tool use has long been a hallmark of human intelligence and a practical problem for solving a vast number of robotic applications. However, machines are still bad at applying the right amount of force to control tools that are not securely attached to the hand.
To more reliably operate the above tools, researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) are working with Toyota Research Institute (TRI) to understand the tools and optimize them for specific tasks. We have designed a system that can apply a large amount of force. , like squeezing out liquids or writing out words with a pen.
system, dubbing Series Elastic End Effector, or SEED, uses a soft bubble gripper and an embedded camera to map how the gripper deforms in 6-dimensional space (think airbag inflation and deflation) and apply forces to the tool. add. With six degrees of freedom, you can move an object left/right, up/down, forward/backward, roll, pitch, and yaw. A closed-loop controller (a self-adjusting system that maintains a desired state without human intervention) uses SEED and visuo-haptic feedback to adjust the robot arm’s position and apply the desired force.
This is useful for those using the tool when there is uncertainty in table height, for example, as the pre-programmed trajectory may miss the table entirely. “For what we call a hybrid force-position controller, we have relied heavily on the work of Mason, Raibert, and Craig,” said a Ph.D. student in electrical engineering and computer science at MIT and a CSAIL affiliate. said its leader, Hyung Ju Suh.author A new paper on SEED“The idea is that when you’re writing on a chalkboard and you’re actually moving in three dimensions, you want to be able to control position in some axes and force in others.”
Rigid-body robots and their counterparts can only take us so far. Its softness and suppleness give it luxury and transformability, sensing the interaction of your hand with your tool.
With SEED, every run the robot senses is the latest 3D image from the gripper, so it tracks in real time how the gripper changes shape around the object. These images are used to reconstruct the position of the tool, and the robot uses the learned model to map the position of the tool to the measured force. The trained model is obtained using the robot’s previous experience. This experience interferes with the force-torque sensor to capture the stiffness of the bubble gripper. Now, when the robot senses the force, it compares it to the force commanded by the user and says, “I know the force I’m sensing right now isn’t that great. I need to push harder.” move in the direction of increasing . Everything takes place in 6D space.
During the “squeegee task”, the SEED was given just the right amount of force to wipe off the plane’s liquid, which the baseline method struggled to get a proper sweep. When asked to put a piece of paper to the pen, the bot was able to effectively write “MIT” and apply just the right amount of force to tighten the screw.
SEED was aware of the fact that force or torque had to be commanded for certain tasks, but there was an upper limit to the hardness that could be exerted, as gripping too tightly would inevitably cause the item to slip. Yes, and for rigid robots, you can simulate a system that is softer than its natural mechanical stiffness, but not vice versa.
Currently the system expects a very specific geometry for the tool. It has to be cylindrical, and there are still many limitations on how to generalize when encountering new types of shapes. It may contain any tool that can handle it.
“It wouldn’t surprise anyone that compliance would be useful for a tool, or that force sensing would be a good idea. said Russ Tedrake, co-author of the paper, Toyoda Professor and Principal of MIT’s Electrical and Computer Science, Aerospace Engineering, and Mechanical Engineering. CSAIL investigator. “Here we consider tuning a very soft six-degree-of-freedom stiffness directly at the hand-tool interface and show that there are several advantages to doing so.”
Suh co-wrote the paper with Naveen Kuppuswamy, a senior researcher at the Toyota Research Institute. Tao Pang, Mechanical Engineering PhD student at MIT and his CSAIL affiliates. TRI’s Paul Mitiguy and Alex Alspach. and Tedreake. They plan to present their work at the IEEE/RSJ International Conference on Intelligent Robots and Systems in October.
Toyota Research Institute provided funding to support this work.
#Soft #robot #grips #moderate #force #MIT #News