AI Paper Atlas

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

Why Dedicated Critics: Eliminating Target Drift in Multi-Constraint RL

Yue Yang, Chenghao Huang, Hao Wang

ICML 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Constrained policy optimization

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use dedicated per-constraint critics instead of a mixed cost criticPPO-Lagrangian · SafetyCarGoal1-v0episodic reward: 1.12 -> 14.56Improved
use dedicated per-constraint critics instead of a mixed cost criticPPO-Lagrangian · SafetyCarGoal1-v0episodic cost: 55.34 -> 21.72Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyCarButton1-v0episodic cost: 107.14 -> 51.40Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyAntVelocity-v1episodic reward: 2821.72 -> 3324.67Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyAntVelocity-v1episodic cost: 28.52 -> 13.01Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyHalfCheetahV-v1episodic cost: 45.82 -> 4.14Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyHalfCheetahV-v1episodic reward: 2234.25 -> 3035.76Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyPointGoal1-v0episodic reward: 15.78 -> 13.03Hurt
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyPointGoal1-v0episodic cost: 52.81 -> 23.97Improved
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyHopperVelocity-v1episodic reward: 1238.83 -> 1002.73Hurt
use dedicated critics instead of a mixed criticPPO-Lagrangian · SafetyHopperVelocity-v1episodic cost: 17.21 -> 14.87Improved
use dedicated per-constraint critics instead of a mixed cost criticPPO-Lagrangian · Power system, 5 constraintsconstraint violationImproved

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