Research map / Vision-language and multimodal / Robot action and navigation
WholeBodyVLA: Towards Unified Latent VLA for Whole-body Loco-manipulation Control
Haoran Jiang, Jin Chen, Qingwen Bu, Li Chen, Modi Shi, Yanjie Zhang, Delong Li, Chuanzhe Suo, wang chuang, zhihui peng, Hongyang Li
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 latent action model pretraining on human videos (ablation) | WholeBodyVLA · Real-robot suites avg | success rate: 39.3% -> 78.0%; +38.7% | Improved |
| add LMO discrete-command RL policy instead of velocity-based RL (ablation) | WholeBodyVLA · Real-robot suites avg | success rate: 54.0% -> 78.0%; 24% lower | Improved |
| add locomotion video pretraining to manipulation-only LAM (ablation) | WholeBodyVLA · Real-robot suites avg | success rate: 63.3% -> 78.0%; +14.7% | Improved |
| use separate locomotion and manipulation LAMs instead of shared LAM (ablation) | WholeBodyVLA · Real-robot suites avg | success rate: 66.0% -> 78.0% | Improved |
| add directional accuracy reward to LMO (ablation) | LMO · MuJoCo turning | orientation error: 0.28 -> 0.19 | Improved |
| add stage 2 refinement to LMO (ablation) | LMO · MuJoCo turning | position error: 0.20 -> 0.05 | Improved |
| add stage 2 refinement to LMO (ablation) | LMO · MuJoCo squatting | CoM sway: 0.07 -> 0.03 | Improved |
| add stage 1 minimal gait training to LMO (ablation) | LMO · MuJoCo turning | position error: 0.46 -> 0.05 | Improved |
| replace velocity-based RL with LMO (ablation) | RL controller · MuJoCo turning | position error: 0.26 -> 0.05 | Improved |
| replace velocity-based RL with LMO | RL controller · MuJoCo forward&backward | position error: 0.24 -> 0.21 | No clear effect |
| use more human egocentric video pretraining (0 to 100%) | WholeBodyVLA · locomotion generalization | avg success rate | Improved |
| use more AgibotWorld data for latent learning | WholeBodyVLA · manipulation generalization | avg success rate | Improved |
Closest papers by meaning
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.91)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.90)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · similarity 0.89)
- Hierarchical World Models as Visual Whole-Body Humanoid Controllers (ICLR 2025 · similarity 0.89)
- HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning (NeurIPS 2026 · similarity 0.89)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.88)
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.
