AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Multi-Objective Reinforcement Learning Using Routed Ensembles and Trajectory Attribution

Zhenshuo Zhang, Ruoxuan Xiong, Hongyang Zhang

NeurIPS 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Tree search and policy optimization · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a multi-objective RL approach using routed ensembles and trajectory attribution to learn policies that balance competing objectives across preference weights.

Problem

Multi-objective reinforcement learning must learn policies that balance competing objectives, and optimal policies can vary sharply with the preferred objective weights.

Approach

Uses routed ensembles with trajectory attribution to handle sharp variation of optimal policies across objective weights; the abstract is truncated so the specific mechanism is only partly stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
route trajectories to basis policies by highest scalarized returnmulti-task RL base learner · Meta-World MT10success rate (%): 82.0 -> 85.6Improved
route trajectories to the best basis policy rather than assigning uniformly at random (ablation)routed ensemble · Meta-World MT10success rate (%): 81.3 -> 85.6Improved
route trajectories instead of uniform assignment (ablation)routed ensemble · MO-HalfCheetah2dHV (x10^6): 3.17 -> 5.17Improved
route trajectories instead of uniform assignment (ablation)routed ensemble · MO-HopperHV (x10^7): 1.56 -> 1.83Improved
route trajectories instead of uniform assignment (ablation)routed ensemble · Fruit-TreeHV (x10^3): 8.29 -> 9.30Improved
route trajectories instead of uniform assignment (ablation)routed ensemble · RLHF, Qwen3-1.7B HelpSteer2hypervolume: 0.47 -> 0.54Improved
route trajectories instead of uniform assignment (ablation)routed ensemble · RLHF, Llama-3.2-1B HelpSteer2hypervolume: 0.35 -> 0.35No clear effect
use a routed ensemble instead of a single policyrouted ensemble · synthetic RLHF environmenthypervolume: 0.48 -> 0.56Improved
add routing across an ensemble of basis policiespolicy ensemble · synthetic environmentcoverability (chi^2 divergence): 1.3 -> 0.2Improved
increase number of basis policies N from 2 to 5 (ablation)routed ensemble · MO-Gymnasium MuJoCoHV: best N=3 for HalfCheetah (5.21) and Hopper (1.87), N=5 for Ant (2.12)Mixed
use nonlinear (quadratic, cubic, RBF) kernel instead of linear kernel (ablation)kernel surrogate attribution · synthetic datasetfitting loss: 0.2 -> 0.1Improved
use kernel surrogate instead of linear surrogates for attributiontrajectory attribution · Synergy MTRLLDS: 0.16 -> 0.91Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy