Research map / Reinforcement learning / Visual and representation RL
Leave no Trace: Learning to Reset for Safe and Autonomous Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Jointly learn a forward policy and a reset policy instead of using forward-only learning | Ball in Cup · Ball in Cup | task reward | Helped |
| Use a learned reset policy and early aborts instead of resetting after every episode | Leave No Trace · Pusher | hard resets: half the number of hard resets | Helped |
| Use a learned reset policy and early aborts instead of resetting after every episode | Leave No Trace · Cliff Cheetah | hard resets: an order of magnitude fewer hard resets | Helped |
| Increase the early-abort threshold from 0 to 0.4 (ablation) | Leave No Trace · Gridworld | hard resets: 78% | Helped |
| Increase the early-abort threshold from 0 to 0.4 (ablation) | Leave No Trace · Gridworld | steps to solve | No clear effect |
| Increase the early-abort threshold to 80 (ablation) | Leave No Trace · Pusher | hard resets: 33% | Helped |
| Decrease the early-abort threshold (ablation) | Leave No Trace · Pusher and Cliff Cheetah | hard resets | Hurt |
| Use an early-abort threshold of 10 (ablation) | Leave No Trace · Cliff Cheetah | early aborts: 69% | Helped |
| Increase reset attempts from 1 to 4 (ablation) | Leave No Trace · Pusher | hard resets: 2.5x | Helped |
| Increase reset attempts from 1 to 4 (ablation) | Leave No Trace · Pusher | reward: less than 25% | Hurt |
| Increase reset attempts (ablation) | Leave No Trace · Cliff Cheetah | hard resets: nearly 0 | Helped |
| Increase ensemble size from 1 to 50 (ablation) | Leave No Trace · Pusher | task reward | Helped |
Closest papers by meaning
- A Provable Approach for End-to-End Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Simplifying Constraint Inference with Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.87)
- Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning (AAAI 2025 · similarity 0.87)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.87)
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.