Reproduce “Multi-Objective Reinforcement Learning Using Routed Ensembles and Trajectory Attribution” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhenshuo Zhang, Ruoxuan Xiong, Hongyang R. Zhang
The authors introduce a routed ensemble for multi-objective reinforcement learning, assigning each trajectory to the candidate policy that scores it best for its preference and updating policies on their assigned data. In synthetic tests, routing yields lower cumulative regret; on Meta-World MT10, it reaches 85.6 ± 1.8% success rate, versus 82.0 ± 1.1% for Soft Modularization. These findings are limited to the reported settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.