AI Paper Atlas

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

Enhancing Safe Exploration Using Safety State Augmentation

Aivar Sootla, Alexander Imani Cowen-Rivers, Jun Wang, Haitham Bou Ammar

NeurIPS 2022 · Reinforcement learning › Policy gradient and actor-critic · approach: Safety layers and shielding · 8 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 safety state augmentation to L-PPO (ablation)L-PPO · Safe pendulum swing-upconstraint violations / convergenceImproved
Add safety state augmentation to PID-Lagrangian (ablation)PID-Lagrangian · Static point goal, safety budgets d=1,10,20return and safety cost stabilityImproved
Add safety budget scheduling (simmering) to PID-L without state augmentation (ablation)PID-L · Static point goaltraining stabilityImproved
Combine state augmentation with simmering (ablation)PID-L · Static point goalreturn and safetyImproved
Simmer PID-L versus PID-LPID-L · Safety Gym Point and Car tasksreturnNo clear effect
Simmer PID-L versus PID-LPID-L · Safety Gym Point and Car taskssafety cost and cost rateImproved
Use PI-Simmer or Q-Simmer instead of PO-PPO alonePO-PPO · Probability-one constraints, trajectories with violationsnumber of violating trajectoriesImproved
Use PI Simmer rather than Q SimmerSimmer RL · Probability-one constraintsviolationsImproved
Apply PI/Q Simmer to average-constrained RL with state augmentationL-PPO w SA · Pendulum, static point goalviolationsNo clear effect

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