A new arXiv paper, "Towards Agentic Robots: Model, Data, Environment, and Harness," argues that foundation models are driving a paradigm shift in robotic intelligence. The authors describe a move away from low-level action prediction toward a more holistic, goal-directed orchestration of the entire robotic workflow.

According to the abstract, this orchestration spans data collection, model training, task planning, and real-world execution. The paper's title suggests that these four elements—model, data, environment, and harness—form the key pillars of the proposed framework for building agentic robots.

The source is a single arXiv abstract, so details on methodology and results are not yet available. The significance lies in the framing: robots are no longer just predicting actions but are being positioned as autonomous agents that manage their own learning and execution pipelines.