Ken Goldberg and Team Published on BAIR

03 Nov, 2017

Ken Goldberg and Team Published on BAIR

The Berkeley Artificial Intelligence Research (BAIR) Blog published an article about Ken Goldberg and his team's work on an Off-Policy Imitation Learning (IL) algorithm called DART. Within IL, also known as Learning from Demonstration (LfD), Off-Policy (Behavior Cloning) is when demonstrations are given independent of the robot's policy.

Goldberg, PhD student Michael Laskey, and EECS undergraduate student Jonathan Lee worked to bolster Off-Policy Imitation Learning by "injecting noise into the supervisor’s actions [to] improve robustness [in a process called DART]. The injected noise allows the supervisor to provide corrective examples for the type of errors the trained robot is likely to make. However, because the optimized noise is small, it alleviates the difficulties of On-Policy methods."

They further write,

We evaluate DART in simulation with an algorithmic supervisor on MuJoCo tasks (Walker, Humanoid, Hopper, Half-Cheetah) and physical experiments with human supervisors training a Toyota HSR robot to perform grasping in clutter, where a robot must search through clutter for a goal object. Finally, we show how DART can be applied in a complex system that leverages both classical robotics and learning techniques to teach the first robot to make a bed.

Goldberg, Laskey, and Lee will be presenting their findings at the 1st Annual Conference on Robot Learning (CoRL 2017), which will take place in Mountain View, CA, on November 13-15, 2017.

Read more about their work with DART on BAIR here.

Browse their codebase on GitHub here.