Research map / Vision-language and multimodal / Robot action and navigation
HALO: A Unified Vision-Language-Action Model for Embodied Multimodal Chain-of-Thought Reasoning
Quanxin Shou, Fangqi Zhu, Shuang Chen, Puxin Yan, Zhengyang Yan, Yikun Miao, Xiaoyi Pang, Zicong Hong, Ruikai Shi, Hao HUANG, Jie ZHANG, Song Guo
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 EM-CoT (text reasoning plus visual subgoal) to HALO (ablation) | HALO · RoboTwin 2.0 Easy | avg success rate (%): 75.3 -> 80.5 | Improved |
| add EM-CoT (text reasoning plus visual subgoal) to HALO (ablation) | HALO · RoboTwin 2.0 Hard | avg success rate (%): 21.2 -> 26.4 | Improved |
| add textual reasoning to EM-CoT (ablation) | HALO · RoboTwin 2.0 Hard | avg success rate (%): 18.3 -> 26.4 | Improved |
| add textual reasoning to EM-CoT (ablation) | HALO · RoboTwin 2.0 Easy | avg success rate (%): 77.8 -> 80.5 | Improved |
| add visual subgoal images to EM-CoT (ablation) | HALO · RoboTwin 2.0 Hard | avg success rate (%): 22.5 -> 26.4 | Improved |
| add visual subgoal images to EM-CoT (ablation) | HALO · RoboTwin 2.0 Easy | avg success rate (%): 76.1 -> 80.5 | Improved |
| add visual generation data to pretraining (ablation) | HALO pretraining · RoboTwin 2.0 Hard | avg success rate (%): 10.5 -> 21.2 | Improved |
| add visual generation data to pretraining (ablation) | HALO pretraining · RoboTwin 2.0 Easy | avg success rate (%): 58.2 -> 75.3 | Improved |
| add textual VQA data to pretraining (ablation) | HALO pretraining · RoboTwin 2.0 Easy | avg success rate (%): 42.9 -> 58.2 | Improved |
| add textual VQA data to pretraining (ablation) | HALO pretraining · RoboTwin 2.0 Hard | avg success rate (%): 3.9 -> 10.5 | Improved |
| add action prediction pretraining data (ablation) | HALO pretraining · RoboTwin 2.0 Hard | avg success rate (%): 0.0 -> 3.9 | Improved |
| add action prediction pretraining data (ablation) | HALO pretraining · RoboTwin 2.0 Easy | avg success rate (%): 32.4 -> 42.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports HALO as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · better on 8 of 8 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.92)
- Vision-Language-Action Instruction Tuning: From Understanding to Manipulation (ICLR 2026 · similarity 0.91)
- ACoT-VLA: Action Chain-of-Thought for Vision-Language-Action Models (CVPR 2026 · similarity 0.91)
- DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge (NeurIPS 2025 · similarity 0.91)
- CoA-VLA: Improving Vision-Language-Action Models via Visual-Text Chain-of-Affordance (ICCV 2025 · similarity 0.91)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.90)
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.
