Routed ensembles improve learning for tasks with conflicting goals
Multi-Objective Reinforcement Learning Using Routed Ensembles and Trajectory Attribution
Machine LearningArtificial IntelligenceMultiagent SystemsRobotics
Summary
Balancing different goals is hard when teaching AI to make decisions, especially when those goals conflict, like being fast but saving energy. The authors introduced a way to measure how much these goals clash and designed a method that keeps multiple base decision strategies ready. The AI learns by directing tasks to the best strategy for each situation. They also created a way to understand how learning from some experiences affects learning from others, helping to choose better training experiences. Their techniques led to better results in simulated robotics and task environments.
What this means in practice
- •For robotics engineers: Improve robot control systems to better balance competing goals like speed and energy efficiency by using routed ensemble policies.
- •For machine learning engineers: Enhance multi-objective reinforcement learning by applying trajectory attribution to select training experiences that improve policy performance.
Authors
Zhenshuo Zhang, Ruoxuan Xiong, Hongyang R. Zhang
Abstract
Multi-objective reinforcement learning is a fundamental problem for modern AI, with applications spanning post-training, reasoning models, and robotic control. A central challenge is to learn policies that balance competing objectives, such as helpfulness versus honesty in safety and alignment, or speed versus energy in robotics. The optimal policy can vary sharply with objective-weighting combinations, and the extent of this variation depends crucially on the degree of conflict among objectives, making per-preference policy learning sample-inefficient. In this paper, we introduce objective-conflict coefficients and derive near-optimal regret bounds for online settings using this measure, thereby quantifying the impact of misalignment on learning efficiency. Then, we design routed ensemble policy learning, an algorithm that maintains a small set of basis policies and dynamically routes trajectories to the policy with the highest potential reward. Finally, we consider trajectory selection as another approach to mitigate objective conflict, and design a kernel surrogate trajectory attribution algorithm that estimates how feedback signals from a set of trajectories influence policy learning on other trajectories. The kernel can capture nonlinear interactions such as synergy and antagonism between trajectories. We evaluate our approach across diverse environments and benchmarks: routed ensembles improve average performance by 4.3% over baselines for online environments from Meta-World and MO-Gymnasium, and kernel surrogate attribution improves correlation with leave-one-out analysis over existing baselines by 50%. When applied to trajectory selection, kernel-based attribution yields a 27% improvement in reward over baseline attribution methods.