Reinforcement learning for non-prehensile manipulation: Transfer from simulation to physical system

TL;DR

This paper demonstrates that control policies trained via model-based reinforcement learning in simulation can effectively transfer to real-world robotic systems, with ensemble training enhancing robustness against modeling errors.

Contribution

It shows successful transfer of simulation-trained policies to physical robots using a modified natural policy gradient and ensemble modeling for robustness.

Findings

Policies trained in simulation work on real robots without retraining.

Ensemble of models improves robustness to simulation-reality mismatch.

Control policies effectively push objects to target positions.

Abstract

Reinforcement learning has emerged as a promising methodology for training robot controllers. However, most results have been limited to simulation due to the need for a large number of samples and the lack of automated-yet-safe data collection methods. Model-based reinforcement learning methods provide an avenue to circumvent these challenges, but the traditional concern has been the mismatch between the simulator and the real world. Here, we show that control policies learned in simulation can successfully transfer to a physical system, composed of three Phantom robots pushing an object to various desired target positions. We use a modified form of the natural policy gradient algorithm for learning, applied to a carefully identified simulation model. The resulting policies, trained entirely in simulation, work well on the physical system without additional training. In addition, we show that training with an ensemble of models makes the learned policies more robust to modeling errors, thus compensating for difficulties in system identification.