Inverse Reinforcement Learning with Multiple Planning Horizons

TL;DR

This paper addresses inverse reinforcement learning where experts plan with different horizons and unknown discount factors, proposing algorithms to learn a shared reward function and individual discount factors to accurately replicate expert policies.

Contribution

The authors develop algorithms to learn a global reward function with agent-specific discount factors, handling unknown planning horizons in IRL.

Findings

Algorithms successfully recover reward functions across multiple domains.

The approach generalizes well to different planning horizons.

Feasible solution space characterized for reward and discount factors.

Abstract

In this work, we study an inverse reinforcement learning (IRL) problem where the experts are planning under a shared reward function but with different, unknown planning horizons. Without the knowledge of discount factors, the reward function has a larger feasible solution set, which makes it harder for existing IRL approaches to identify a reward function. To overcome this challenge, we develop algorithms that can learn a global multi-agent reward function with agent-specific discount factors that reconstruct the expert policies. We characterize the feasible solution space of the reward function and discount factors for both algorithms and demonstrate the generalizability of the learned reward function across multiple domains.