Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Thinking in 360deg: Humanoid Visual Search in the Wild
Heyang Yu, Yinan Han, Xiangyu Zhang, Baiqiao Yin, Bowen Chang, Xiangyu Han, Xinhao Liu, Jing Zhang, Marco Pavone, Chen Feng, Saining Xie, Yiming 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 SFT on multi-turn trajectories (ablation) | Qwen2.5-VL-3B · H*Bench object search | success rate: 14.83 -> 40.83 | Improved |
| add SFT on multi-turn trajectories (ablation) | Qwen2.5-VL-3B · H*Bench path search | success rate: 6.44 -> 23.00 | Improved |
| add GRPO RL after SFT (ablation) | HVS-3B · H*Bench object search | success rate: 40.83 -> 47.38 | Improved |
| add GRPO RL after SFT (ablation) | HVS-3B · H*Bench path search | success rate: 23.00 -> 24.94 | Improved |
| add GRPO RL after SFT (ablation) | HVS-3B · H*Bench path search medium | success rate: 23.03 -> 20.18 | Hurt |
| add GRPO RL after SFT (ablation) | HVS-3B · H*Bench path search extreme | success rate: 14.81 -> 12.04 | Hurt |
| apply RL directly without SFT (ablation) | HVS-3B · H*Bench | instruction following | Hurt |
| train on object search, test on path search (ablation) | HVS-3B · H*Bench path search | success rate: 6.4 -> 20.7 | Improved |
| train on path search, test on object search (ablation) | HVS-3B · H*Bench object search | success rate: 14.8 -> 29.5 | Improved |
| add distance-to-goal reward in GRPO (ablation) | HVS-3B · H*Bench path search | success rate: 23.44 -> 21.37 | Hurt |
| use shorter GRPO rollouts (ablation) | HVS-3B · H*Bench | success rate | No clear effect |
| use passive panorama instead of active rotation (ablation) | HVS-3B · H*Bench (Gemma-3-4B-it) | success rate | Hurt |
Closest papers by meaning
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots (ICLR 2024 · similarity 0.87)
- FlySearch: Exploring how vision-language models explore (NeurIPS 2025 · similarity 0.87)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.86)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.85)
- Habitat Synthetic Scenes Dataset (HSSD-200): An Analysis of 3D Scene Scale and Realism Tradeoffs for ObjectGoal Navigation (CVPR 2024 · similarity 0.85)
- SenseSearch: Empowering Vision-Language Models with High-Resolution Agentic Search-Reasoning via Reinforcement Learning (CVPR 2026 · similarity 0.85)
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.
