Reward Learning using Structural Motifs in Inverse Reinforcement Learning
This addresses inefficiencies in inverse reinforcement learning for domains like robotics and health, though it appears incremental as it builds on existing IRL methods with structural enhancements.
The paper tackled the problem of learning reward functions from expert trajectories in long-horizon, complex sequential tasks by proposing SMIRL, a method that uses structural motifs from finite-state-automata, and showed it successfully learned all four complex tasks where baselines failed and improved sample efficiency on a simpler task.
The Inverse Reinforcement Learning (\textit{IRL}) problem has seen rapid evolution in the past few years, with important applications in domains like robotics, cognition, and health. In this work, we explore the inefficacy of current IRL methods in learning an agent's reward function from expert trajectories depicting long-horizon, complex sequential tasks. We hypothesize that imbuing IRL models with structural motifs capturing underlying tasks can enable and enhance their performance. Subsequently, we propose a novel IRL method, SMIRL, that first learns the (approximate) structure of a task as a finite-state-automaton (FSA), then uses the structural motif to solve the IRL problem. We test our model on both discrete grid world and high-dimensional continuous domain environments. We empirically show that our proposed approach successfully learns all four complex tasks, where two foundational IRL baselines fail. Our model also outperforms the baselines in sample efficiency on a simpler toy task. We further show promising test results in a modified continuous domain on tasks with compositional reward functions.