Bayesian Gaussian mixture model for robotic policy imitation
This work addresses the challenge of ensuring safe and appropriate robot behavior in unknown states during imitation learning, which is incremental as it builds on existing imitation methods with uncertainty quantification.
The paper tackles the problem of robotic policy imitation where small errors cause the robot to leave demonstrated states, by proposing a Bayesian method to quantify action uncertainty and fuse imitation with additional policies. The approach is validated on a Panda robot for three manipulation tasks in continuous domains.
A common approach to learn robotic skills is to imitate a demonstrated policy. Due to the compounding of small errors and perturbations, this approach may let the robot leave the states in which the demonstrations were provided. This requires the consideration of additional strategies to guarantee that the robot will behave appropriately when facing unknown states. We propose to use a Bayesian method to quantify the action uncertainty at each state. The proposed Bayesian method is simple to set up, computationally efficient, and can adapt to a wide range of problems. Our approach exploits the estimated uncertainty to fuse the imitation policy with additional policies. It is validated on a Panda robot with the imitation of three manipulation tasks in the continuous domain using different control input/state pairs.