The release of GPT-6 Astra introduces agentic robotics, leveraging foundation models and embodied harnesses for sustained physical autonomy.
The release of GPT-6 Astra has catalyzed a transformative shift in robotics, introducing the concept of agentic robotics. This new paradigm, detailed in a research paper published on arXiv on November 13, 2026, by Grid Robotics and lead authors Qingfeng Lan, Duo Li, Siyao Li, and others, emphasizes the integration of foundation models with embodied harnesses to achieve reliable physical autonomy. The paper has garnered significant attention, with 107 community upvotes on HuggingFace Daily Papers as of November 16, 2026.
At the heart of agentic robotics is the Embodied System 2 framework, which combines a foundation model with an embodied harness to form the decision-making core. Unlike traditional robotics systems that rely on task-specific policies, agentic robotics leverages a foundation model to handle open-world embodied tasks. This system operates in a closed loop, observing the environment, interpreting user intent, reasoning about tasks, invoking necessary skills and tools, verifying outcomes, and adapting strategies as needed.
The research highlights the challenges of translating general-purpose reasoning into effective and verifiable actions under physical constraints, such as limited reaction times, mechanical limitations, and unpredictable environmental changes. These constraints underscore the importance of developing a shared evaluation framework and research roadmap to guide advancements in the field.
The authors propose an evaluation framework that measures both the capabilities of the foundation model and the performance of the closed-loop system. This framework is designed to address the lack of a unified definition and reliable benchmarks for agentic robotics.
Source link







