Research map / Reinforcement learning / Policy gradient and actor-critic
Stochastic Minimum-Cost Reach-Avoid Reinforcement Learning
Jingduo Pan, Taoran Wu, Yiling Xue, Bai Xue
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use RAPCPO instead of RC-PPO under same iteration budget | RAPCPO · PointGoal | reach rate: 62.29% -> 78.49% | Improved |
| use RAPCPO instead of RC-PPO under same iteration budget | RAPCPO · FixedWing | reach rate: 73.98% -> 88.67% | Improved |
| add compensation factor to enhanced Bellman objective (ablation) | RAPCPO · PointGoal, FixedWing, Safety Hopper, Safety HalfCheetah | additional cumulative cost | Improved |
| add compensation factor to fixed-gamma Bellman estimate (ablation) | reach-avoid probability estimate · FrozenLake | estimate calibration | Improved |
| use enhanced Bellman formulation instead of fixed-gamma (ablation) | RAPCPO · Safety Hopper | reach rate: 0.3218 -> 0.9434 | Improved |
| use enhanced Bellman formulation instead of fixed-gamma (ablation) | RAPCPO · PointGoal | reach rate: 0.4526 -> 0.7821 | Improved |
| set p to 0 (ablation) | RAPCPO · Safety Hopper | reach rate | Hurt |
| vary p within 0.1 to 0.7 (ablation) | RAPCPO · Safety Hopper | reach rate and cost | Improved |
| raise p to 0.8-1.0 (ablation) | RAPCPO · Safety Hopper | additional cumulative cost | Hurt |
| use CVaR-constrained methods (Saute, CPPO) with state-wise costs | Saute, CPPO · deterministic and stochastic MuJoCo tasks | cost and reach rate | Hurt |
| use expectation-based constraints instead of CVaR constraints | CMDP baselines · deterministic and stochastic MuJoCo tasks | cost and reach rate | Improved |
Closest papers by meaning
- Solving Minimum-Cost Reach Avoid using Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Absolute State-wise Constrained Policy Optimization: High-Probability State-wise Constraints Satisfaction (NeurIPS 2026 · similarity 0.87)
- Reinforcement Learning with Random Time Horizons (ICML 2025 · similarity 0.87)
- Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively (AAAI 2024 · similarity 0.87)
- Efficient Exploration in Average-Reward Constrained Reinforcement Learning: Achieving Near-Optimal Regret With Posterior Sampling (ICML 2024 · similarity 0.86)
- Polynomial-Time Approximability of Constrained Reinforcement Learning (ICML 2025 · similarity 0.86)
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.
