# Bayesian Gaussian mixture model for robotic policy imitation

## TL;DR

This paper introduces a Bayesian Gaussian mixture model to quantify action uncertainty in robotic policy imitation, enhancing the robot's ability to handle unknown states by fusing policies based on uncertainty estimates.

## Contribution

It presents a simple, efficient Bayesian approach for uncertainty estimation in imitation learning, enabling better policy fusion and robustness in robotic manipulation tasks.

## Key findings

- Effective uncertainty quantification improves policy robustness.
- Successful validation on a Panda robot across multiple manipulation tasks.
- Fused policies outperform standard imitation in unknown states.

## Abstract

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.