Event
Multi-Objective Reinforcement Learning Routing Ensemble and Trajectory Attribution Method
1 reports · 1 sources
Zhenshuo Zhang, Ruoxuan Xiong, and Hongyang R. Zhang propose a routing ensemble policy learning algorithm for multi-objective reinforcement learning: it maintains a small set of base policies and dynamically routes trajectories to the policy with the highest potential reward. The authors also propose a kernel surrogate trajectory attribution algorithm to estimate the influence of trajectory feedback signals on the policy learning of other trajectories. The paper introduces a goal conflict coefficient and, based on it, derives a near-optimal regret bound for the online setting to quantify the impact of objective misalignment on learning efficiency. The paper states it will be published at NeurIPS 2026 (extended version).
Report timeline
- Multi-Objective Reinforcement Learning with Routed Ensembles and Trajectory Attribution
This paper studies policy learning in multi-objective reinforcement learning, proposing objective-conflict coefficients and using them to derive near-optimal regret bounds for online settings, quantifying the impact of objective misalignment on learning efficiency. The authors design a routed ensemble policy learning algorithm that maintains a small set of basis policies and dynamically routes trajectories to the policy with the highest potential reward; they also propose a kernel surrogate trajectory attribution algorithm that estimates how trajectory feedback signals affect the policy learning of other trajectories.
arXiv cs.RO ·