Workshop on Agentic Robotics

Across the broader AI field, agents increasingly operate by decomposing tasks, generating code, calling tools, managing memory, and revising plans based on intermediate feedback. This pattern is spreading to robotics, where reasoning, skills, and verification are composed at inference time on top of learned policies. Bringing the paradigm onto physical robots, however, surfaces challenges absent from digital settings: actions are continuous and irreversible, observations are noisy and partial, and decisions must be made in real time. We call this emerging area the study of agentic robotics.

Research on agentic robotics is currently distributed across several communities — foundation-model and VLA researchers building reasoning loops on learned policies, planning researchers using language models as abstractions, control and policy-learning researchers studying long-horizon execution, and software-agent researchers extending their methods to embodied tasks. These efforts share substantial common ground but rarely meet in the same session. This workshop creates that forum, with the explicit aim of surfacing shared open problems and points of disagreement across these communities.