Self-Supervised Correspondence in Visuomotor Policy Learning
A video of the latest research by Peter Florence, Lucas Manuelli, Russ Tedrake of MIT’s CSAIL lab. It was republished by MIT.
Having robots learn dexterous tasks requiring real-time hand-eye coordination is hard. Many tasks that we would consider simple, like hanging up a baseball cap on a rack, would be very challenging for most robot software. What’s more, for a robot to learn each new task, it typically takes significant amounts of engineering time to program the robot. Pete Florence and Lucas Manuelli in the Robot Locomotion Group took a step closer to that goal with their work. Their paper will be presented at International Conference on Robotics and Automation (ICRA) in May in Paris.
Link to the paper: https://arxiv.org/abs/1909.06933