3 ms·
Not teleoperating can have certain disadvantages due to mismatches between how humans move vs. how robots move though. See here: https://evjang.com/2024/08/31/m
by v9v 11mo ago
Not teleoperating can have certain disadvantages due to mismatches between how humans move vs. how robots move though. See here: https://evjang.com/2024/08/31/motors.html https://evjang.com/2024/08/31/motors.html
- ACCount37 11mo agoIntuitively, yes. But is it really true in practice? Thinking about it, I'm reminded of various "additive training" tricks. Teach an AI to do A, and then to do B, and it might just generalize that to doing A+B with no extra training. Works often enough on things like LLMs. In this case, we use non-robot data to teach an AI how to do diverse tasks, and robot-specific data (real or sim) to teach an AI how to operate a robot body. Which might generalize well enough to "doing diverse tasks through a robot body".
- blueblisters 11mo agoThe exoskeletons are instrumented to match the kinematics and sensor suite of the actual robot gripper. You can trivially train a model on human collected gripper data and replay it on the robot.
- v9v 11mo agoYou mentioned UMI, which to my knowledge runs VSLAM on camera+IMU data to estimate the gripper pose and no exoskeletons are involved. See here: https://umi-gripper.github.io/ https://umi-gripper.github.io/
- blueblisters 11mo ago"Exoskeleton" was inspired by the more recent Dex-Op paper (https://dex-op.github.io/ https://dex-op.github.io/) Calling UMI an "exoskeleton" might be a stretch but the principle is the same - humans use a kinematically matched instrumented end affector to collect data that can be trivially replayed on the robot.