Vision Language Actions using reasoning as action tokens.
Abstract
Vision–Language–Action (VLA) models use reasoning steps (e.g., plans, chains of thought, subgoals, or symbolic decisions) as action tokens, so the model first generates explicit intermediate reasoning that guides downstream control. These reasoning tokens bridge high-level intent and low-level actions, improving interpretability, long-horizon planning, and task generalization across environments.
Fields of Study & Subjects
Robot Manipulation and Learning
AI-based Problem Solving and Planning
Multimodal Machine Learning Applications
Publication details
- Journal / source
- Open Research Knowledge Graph (ORKG)
- Publication date
- 4 February 2026
- Language
- EN
- Document type
- Dataset