Marginalized Importance Sampling for Off-Environment Policy Evaluation

TL;DR

This paper introduces a novel marginalized importance sampling method that combines simulation and offline data to accurately evaluate policies before real-world deployment, addressing key challenges in density ratio estimation.

Contribution

It proposes a two-step density ratio learning approach using occupancy in the simulator, improving efficiency and robustness in off-environment policy evaluation.

Findings

Method generalizes well across different Sim2Sim gaps

Achieves accurate policy evaluation in various environments

Demonstrates successful transfer to real-world robotic arm validation

Abstract

Reinforcement Learning (RL) methods are typically sample-inefficient, making it challenging to train and deploy RL-policies in real world robots. Even a robust policy trained in simulation requires a real-world deployment to assess their performance. This paper proposes a new approach to evaluate the real-world performance of agent policies prior to deploying them in the real world. Our approach incorporates a simulator along with real-world offline data to evaluate the performance of any policy using the framework of Marginalized Importance Sampling (MIS). Existing MIS methods face two challenges: (1) large density ratios that deviate from a reasonable range and (2) indirect supervision, where the ratio needs to be inferred indirectly, thus exacerbating estimation error. Our approach addresses these challenges by introducing the target policy's occupancy in the simulator as an intermediate variable and learning the density ratio as the product of two terms that can be learned separately. The first term is learned with direct supervision and the second term has a small magnitude, thus making it computationally efficient. We analyze the sample complexity as well as error propagation of our two step-procedure. Furthermore, we empirically evaluate our approach on Sim2Sim environments such as Cartpole, Reacher, and Half-Cheetah. Our results show that our method generalizes well across a variety of Sim2Sim gap, target policies and offline data collection policies. We also demonstrate the performance of our algorithm on a Sim2Real task of validating the performance of a 7 DoF robotic arm using offline data along with the Gazebo simulator.