Paper Proposes Agentic Robotics Roadmap
- •Paper proposes an evaluation framework and research roadmap for agentic robotics
- •Embodied System 2 combines a foundation model with an embodied harness
- •Roadmap targets sustained autonomy under physical and safety constraints
A paper published on November 13 proposes a shared evaluation framework and research roadmap for agentic robotics, a robot approach built around a foundation model and an embodied harness. The authors argue that this combination, which they call Embodied System 2, is the decision-making core of agentic robotics.
The paper describes systems that observe their surroundings, interpret user intent, reason about tasks, use skills and tools, check results, and adjust their strategy in a closed loop. Rather than training a separate policy for each task family, these systems apply a foundation model to open-world embodied tasks. The authors say the field still lacks a shared definition, reliable evaluation framework, and research roadmap.
The proposed evaluation measures both model capabilities and closed-loop system performance. The authors identify principles that transfer from digital agents to physical agents, along with challenges to reliable physical autonomy. Robots must turn general reasoning into effective, verifiable actions despite limited reaction time, mechanical constraints, changing environments, and irreversible consequences.
The proposed roadmap aims at sustained autonomy: safely and reliably taking on progressively broader responsibility, from carrying out individual primitives to fulfilling standing roles.
The fastest, most reliable AI news, straight to your inbox