Research map / Reinforcement learning / Visual and representation RL
Emergent Dexterity Via Diverse Resets and Large-Scale Reinforcement Learning
Patrick Yin, Tyler Westenbroek, Zhengyu Zhang, Ignacio Dagnino, Eeshani Shilamkar, Numfor Mbiziwo-Tiapo, Simran Bagaria, Xinlei Liu, Galen Mullins, Andrey Kolobov, Abhishek Gupta
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 OmniReset distilled RGB policy instead of real-only BC (100 demos) | real-world policy · Peg Insertion real | success rate: 2.44% -> 85.37% | Improved |
| use OmniReset distilled RGB policy instead of real-only BC (100 demos) | real-world policy · Leg Twisting real | success rate: 1.82% -> 56.36% | Improved |
| use OmniReset distilled RGB policy instead of real-only BC (100 demos) | real-world policy · Drawer Insertion real | success rate: 1.92% -> 15.38% | Improved |
| distill state-based RL expert into RGB policy | OmniReset · Peg Insertion sim | success rate: 94.97% -> 55.45% | Hurt |
| distill state-based RL expert into RGB policy | OmniReset · Leg Twisting sim | success rate: 93.94% -> 43.69% | Hurt |
| distill state-based RL expert into RGB policy | OmniReset · Drawer Insertion sim | success rate: 89.60% -> 84.00% | Hurt |
| increase number of parallel environments (ablation) | OmniReset PPO · Leg Twisting Hard | success rate from Reaching resets | Improved |
| broaden grasp sampling range (narrow to broad) (ablation) | OmniReset · Leg Twisting | sample efficiency and success rate | Improved |
| use OmniReset instead of Demo Curriculum, BC-PPO, DeepMimic | RL training · Hard task variants | success rate | Improved |
| scale distillation data from 10K to 80K trajectories (ablation) | distilled RGB policy · real-world transfer | success rate | Improved |
| freeze visual encoder instead of end-to-end training (ablation) | distilled RGB policy · distillation | success rate | Hurt |
| replace OSC with joint-space PD control (ablation) | RL and distillation · controller design | success rate | Hurt |
Closest papers by meaning
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 2025 · similarity 0.91)
- Rapidly Adapting Policies to the Real-World via Simulation-Guided Fine-Tuning (ICLR 2025 · similarity 0.90)
- ASID: Active Exploration for System Identification in Robotic Manipulation (ICLR 2024 · similarity 0.88)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.87)
- GENMANIP: LLM-driven Simulation for Generalizable Instruction-Following Manipulation (CVPR 2025 · similarity 0.87)
- DexScale: Automating Data Scaling for Sim2Real Generalizable Robot Control (ICML 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.
