AI Paper Atlas

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

SHAPO: Sharpness-Aware Policy Optimization for Safe Exploration

Kaustubh Mani, Yann Pequignot, Vincent Mai, Liam Paull

ICLR 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Safety layers and shielding · 0 citations (OpenAlex)

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
add SHAPO to baseline on-policy algorithms (CPO, PIDLag, CRPO, SauteRL, TRPOLag)on-policy safe RL baselines · Safety Gym and MuJoCo taskssafety-efficiency tradeoffImproved
add SHAPO to PIDLag across initial multipliersPIDLag · PointGoal1 lambda_init sweepreturn vs cost Pareto frontierImproved
add SHAPO to suppress heavy cost tailsbaselines · Ant-v4, Walker2d-v4episodic cost tail percentileImproved
use Fisher instead of Euclidean perturbations (ablation)SHAPO · Safety Gym and MuJoCo taskssafety and returnImproved
add SAM on critic only (ablation)baseline · Safety Gym and MuJoCo taskscostHurt
add SHAPO on actor only (ablation)baseline · Safety Gym and MuJoCo taskssafetyImproved
add SAM on critic to SHAPO actor (ablation)SHAPO Actor · Safety Gym and MuJoCo taskssafety and efficiencyImproved
use fixed pessimism level instead of fixed delta_Down schedule (ablation)SHAPO · Safety Gym tasksreturn and costNo clear effect
increase delta_Down too large (ablation)SHAPO · hyperparameter sensitivityreturn and costHurt
decrease delta_Down too small (ablation)SHAPO · hyperparameter sensitivityreturn and costNo clear effect
vary rho_critic over wide range (ablation)SHAPO · hyperparameter sensitivityreturn and costNo clear effect
add SHAPO perturbation step (compute overhead)PIDLag · PointGoal1total runtime: 85 min -> 100 minHurt

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