Research map / Vision-language and multimodal / Robot action and navigation
Endowing GPT-4 with a Humanoid Body: Building the Bridge Between Off-the-Shelf VLMs and the Physical World
Yingzhao Jian, Zhongan Wang, Yi Yang, Hehe Fan
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 image labels for VLM visual identification (ablation) | BiBo · Task success, Sit | success rate (%): 48.59 -> 95.84 | Improved |
| add majority voting across five VLM instances (ablation) | BiBo · Task success, Sit | success rate (%): 91.13 -> 95.84 | Improved |
| add IK post-optimization (ablation) | BiBo · Task success, Lift | success rate (%): 6.80 -> 65.42 | Improved |
| condition future motion on actual executed motion (ablation) | BiBo · Task success, Lift | success rate (%): 28.34 -> 65.42 | Improved |
| condition future motion on previously generated motion (ablation) | BiBo · Task success, Composite Hard | success rate (%): 23.81 -> 27.78 | Improved |
| condition on actual executed motion (ablation) | BiBo · HumanML3D | FID (Phy.): 1.414 -> 1.883 | Hurt |
| use latent diffusion model instead of plain diffusion (ablation) | BiBo · HumanML3D | FID: 0.238 -> 0.076 | Improved |
| add causal attention in VAE (ablation) | BiBo · HumanML3D | FID: 0.101 -> 0.076 | Improved |
| use LDM (ablation) | BiBo · Control accuracy MAE head | MAE: 0.0705 -> 0.0310 | Improved |
| condition on previously generated motion (ablation) | BiBo · Motion discontinuity | avg joint acceleration: 0.0698 -> 0.0379 | Improved |
| condition on actual executed motion (ablation) | BiBo · Motion discontinuity | avg joint acceleration: 0.0370 -> 0.0379 | No clear effect |
| use online VLM planning instead of ground-truth plan | BiBo · Task success | success rate (%): Hard 42.86 -> 27.78; Reach 98.91 -> 99.18 | Mixed |
Closest papers by meaning
- An Embodied Generalist Agent in 3D World (ICML 2024 · similarity 0.89)
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · similarity 0.88)
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.88)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.88)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · 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.
