Research map / Reinforcement learning / Visual and representation RL
VIRAL: Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation
Tairan He, Zi Wang, Haoru Xue, Qingwei Ben, Zhengyi Luo, Wenli Xiao, Ye Yuan, Xingye Da, Fernando Castañeda, Shankar Sastry, Changliu Liu, Guanya Shi and 2 more
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 reference state initialization from 200 teleop demos (ablation) | teacher RL · teacher success rate | success rate: below 10% -> nearly 95% | Improved |
| use delta action space instead of absolute joint targets (ablation) | teacher RL · teacher success | success rate | Improved |
| use DINOv3 vision backbone (ablation) | student policy · student success and convergence | success rate | Improved |
| mix student rollouts into distillation (DAgger ratio 0.5) vs pure BC (ablation) | student policy · sim and real deployment | success rate | Improved |
| add history-aware architecture (feed-forward or LSTM) over single-step (ablation) | student policy · student success | success rate | Improved |
| add visual domain randomization (material, dome light, extrinsics) (ablation) | student policy · IsaacSim normalized success | normalized success: 0.649 -> 1.0 | Improved |
| scale teacher training GPUs from 1-2 to 8-16 (ablation) | teacher RL · teacher success | success rate: plateau far below -> above 90% | Improved |
| scale student training GPUs from 1 to 64 (ablation) | student policy · distillation convergence and success | success rate | Improved |
| train grasping on ten objects instead of cylinder only (ablation) | teacher policy · object generalization (normalized success) | success rate | Improved |
| add finger SysID and camera extrinsics alignment (ablation) | sim-to-real · joint position alignment | qualitative | Improved |
Closest papers by meaning
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.90)
- 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)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.87)
- Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation (ICLR 2024 · 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.
