Policy Learning in Spatial Action Spaces for Robotic Manipulation

Explore policy learning in spatial action spaces for robotic manipulation in this Winter 2021 Robotics Colloquium lecture. Delve into the challenges and potential solutions for reinforcement learning in large, complex action spaces corresponding to positions and orientations in SE(2) or SE(3). Examine proposed ideas to facilitate robotic policy learning in spatial action spaces, with evaluations in both simulated and physical systems on challenging manipulation tasks. Learn from Associate Professor Rob Platt of Northeastern University's Khoury College of Computer Sciences, whose expertise spans robotic policy learning, planning, and perception. Gain insights into developing robots capable of performing complex manipulation tasks alongside humans in uncertain environments. The lecture, recorded on February 5, 2021, includes closed captions and runs for 1 hour and 2 minutes.