Research map / Reinforcement learning / Imitation learning
One Demonstration Is Enough for Real-World Robotic Reinforcement Learning
Yuwan Liu, Hongze Yu, song liu, Yuhan Wang, Junge Zhang, Yaodong Yang, Yuanpei Chen, Ceyao Zhang
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 |
|---|---|---|---|
| replace 20-demo SERL with AutoSERL automatic intervention from one demo | SERL · 6 real-robot tasks | success rate: 0/50 -> 50/50 (Plug Insertion) | Improved |
| add sliding window intervention (ablation) | AutoSERL · Plug/USB insertion | training steps to 100% success | Improved |
| add safety recovery mechanism (ablation) | AutoSERL with sliding window · Plug/USB insertion | success rate | Improved |
| add intervention termination (ablation) | AutoSERL · Drawer Opening | success rate | Improved |
| l_term 10 instead of 50 (ablation) | AutoSERL · Drawer Opening | success rate | Improved |
| th1 0.001 m or 0.02 m instead of 0.005 m (ablation) | AutoSERL · Plug Insertion | success rate | Hurt |
| th2 0.005 m or 0.04 m instead of 0.02 m (ablation) | AutoSERL · Plug Insertion | success rate | Hurt |
| use AutoSERL instead of SERL under +-3 cm plug position randomization | SERL · Plug Insertion (positional variation) | success rate | Improved |
| replace human intervention with automatic intervention | HIL-SERL · 6 real-robot tasks | training time to 50/50: 48 -> 33 min (Hanger); 6 -> 8 min (USB) | Mixed |
| use AutoSERL instead of BC / MILES | BC, MILES · 6 real-robot tasks | success rate: e.g. 5/50 and 0/50 -> 50/50 (USB Insertion) | Improved |
Closest papers by meaning
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Policy Decorator: Model-Agnostic Online Refinement for Large Policy Model (ICLR 2025 · similarity 0.86)
- SkiP: When to Skip and When to Refine for Efficient Robot Manipulation (NeurIPS 2026 · similarity 0.86)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 2025 · similarity 0.86)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.86)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model 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.
