Multi-Modal, Multi-Environment Machine Teaching for Robust Reward Learning
2026-07-09 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors study how to teach robots or agents what humans want in a way that works well even when the agents face different situations. They explain that learning from comparisons (which option is better) can give stronger clues about the right goals than other kinds of feedback. They propose a new teaching method that picks different kinds of environments and asks for easy feedback to best figure out goals that work across many settings. Their experiments show this method helps agents learn goals that perform better in new environments compared to simpler approaches.
Autonomous agentsInverse reinforcement learningReward functionsMulti-environment learningFeedback modalitiesMachine teachingMarkov decision processesGeneralizationRegretHierarchical algorithms
Authors
Ali Larian, Qian Lin, Chang Zong Wu, Daniel S. Brown
Abstract
As autonomous agents are increasingly deployed across diverse operational contexts, aligning their behavior with human intent demands reward functions that remain robust to such changes rather than overfitting to any single environment. Inverse reinforcement learning (IRL) provides a principled way to infer such objectives from human feedback. However, existing analyses of optimal teaching approaches for IRL focus on single-environment, demonstration-only settings, leaving underexplored how heterogeneous feedback modalities and environment dynamics jointly constrain reward functions that generalize across multiple environments. Because demonstrations in one MDP entangle reward information with that environments specific structure, the resulting rewards frequently fail to generalize when the agent is deployed in a new setting. We first analyze how different feedback modalities constrain rewards, showing that, in the unlimited-data regime, comparisons impose strictly stronger global constraints than other modalities. Beyond this theoretical analysis, we introduce a hierarchical machine teaching algorithm for reward learning that operates across multiple MDPs. The algorithm first greedily selects informative environments that expose complementary reward constraints, then strategically queries low-cost feedback within those environments. Empirically, our method achieves substantially lower regret and stronger generalization to held-out environments than uniform teaching baselines under identical feedback budgets, demonstrating the importance of multi-environment, multi-modal teaching for learning dynamics-robust reward functions.