MOCI: Multi-Objective Constraint Inference for Reinforcement Learning
A new framework, Multi-Objective Constraint Inference (MOCI), jointly extracts shared constraints and individual preferences from heterogeneous expert trajectories in reinforcement learning. Unlike existing approaches that assume homogeneous demonstrations from a single expert or multiple experts with identical objectives, MOCI models diverse and potentially conflicting behaviors. Empirical evaluations show MOCI significantly outperforms baselines in predictive performance and computational efficiency.
Key facts
- MOCI is a novel framework for constraint inference in reinforcement learning.
- It extracts shared constraints and individual preferences from heterogeneous expert trajectories.
- Existing approaches assume homogeneous demonstrations.
- MOCI handles multiple experts with different objectives.
- Empirical evaluations show MOCI outperforms existing baselines.
- MOCI achieves improved predictive performance.
- MOCI maintains computational efficiency.
- The paper is available on arXiv with ID 2605.06951.
Entities
Institutions
- arXiv