Research map / Reinforcement learning / Visual and representation RL
Opening the Sim-to-Real Door for Humanoid Pixel-to-Action Policy Transfer
Haoru Xue, Tairan He, Zi Wang, Qingwei Ben, Wenli Xiao, Zhengyi Luo, Xingye Da, Fernando Castañeda, Guanya Shi, Shankar Sastry, Linxi Fan, Yuke Zhu
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 |
|---|---|---|---|
| add dome-light randomization with 100% texture randomization (ablation) | DoorMan student · Pull Lever (sim, unseen) | success rate (%): 55.8 -> 80.8 | Improved |
| add dome-light randomization with 100% texture randomization (ablation) | DoorMan student · Push Lever (sim, unseen) | success rate (%): 73.3 -> 85.8 | Improved |
| add dome-light randomization with 100% texture randomization (ablation) | DoorMan student · Push Bar (sim, unseen) | success rate (%): 76.7 -> 85.0 | Improved |
| add dome-light randomization with 10% texture randomization (ablation) | DoorMan student · Pull Lever (sim, unseen) | success rate (%): 50.8 -> 77.5 | Improved |
| add visual randomization (textures and dome light) versus none (ablation) | DoorMan student · Pull Lever (sim, unseen) | success rate (%): 5.0 -> 80.8 | Improved |
| add PBR textures on top of solid-color randomization with dome light (ablation) | DoorMan student · Push Lever (sim, unseen) | success rate (%): 67.5 -> 85.8 | Improved |
| increase texture randomization from 10% to 100% with dome light (ablation) | DoorMan student · Push Lever (sim, unseen) | success rate (%): 79.2 -> 85.8 | Improved |
| increase texture randomization from 10% to 100% with dome light (ablation) | DoorMan student · Push Bar (sim, unseen) | success rate (%): 77.5 -> 85.0 | Improved |
| add GRPO fine-tuning after DAgger distillation (ablation) | DoorMan student · door-opening sub-tasks (sim) | success rate (%): 50-70% -> 80.8-85.8% | Improved |
| increase staged-reset buffer size from 0 to 10 (ablation) | DoorMan teacher · teacher exploration | iterations to reach all stages: fails -> over 4000 iterations | Improved |
| increase staged-reset buffer size from 10 to 100 (ablation) | DoorMan teacher · teacher exploration | iterations to reach all stages: over 4000 -> roughly 1700 iterations | Improved |
| use RL policy instead of expert human teleoperation | DoorMan · real-world door opening | task completion time: 23.8% faster | Improved |
Closest papers by meaning
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.89)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.89)
- Accelerating Visual-Policy Learning through Parallel Differentiable Simulation (NeurIPS 2025 · similarity 0.89)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.87)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · 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.
