Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning

TL;DR

This paper introduces interactive reward functions (IRFs) that enable robots to evaluate and improve their skills through physical interactions, leading to better policy learning and verification without extensive supervision.

Contribution

The authors propose a novel IRF framework that uses example-based interactive behaviors for policy training and verification, reducing the need for detailed reward engineering or demonstrations.

Findings

IRFs improve task performance significantly in simulation and real robot experiments.

IRFs outperform baseline methods with demonstrations or engineered rewards.

IRFs enable effective policy training using only successful outcome examples.

Abstract

Physical interactions can often help reveal information that is not readily apparent. For example, we may tug at a table leg to evaluate whether it is built well, or turn a water bottle upside down to check that it is watertight. We propose to train robots to acquire such interactive behaviors automatically, for the purpose of evaluating the result of an attempted robotic skill execution. These evaluations in turn serve as "interactive reward functions" (IRFs) for training reinforcement learning policies to perform the target skill, such as screwing the table leg tightly. In addition, even after task policies are fully trained, IRFs can serve as verification mechanisms that improve online task execution. For any given task, our IRFs can be conveniently trained using only examples of successful outcomes, and no further specification is needed to train the task policy thereafter. In our evaluations on door locking and weighted block stacking in simulation, and screw tightening on a real robot, IRFs enable large performance improvements, even outperforming baselines with access to demonstrations or carefully engineered rewards. Project website: https://sites.google.com/view/lirf-corl-2022/