🤖 AI Summary
This work addresses the frequent failures of toolchains in robotic policy training and the absence of reliable evaluation and recovery mechanisms. It proposes AgenticRobotics—a backend-agnostic agentic control plane powered by large language models that dynamically orchestrates ephemeral worker nodes to establish a recoverable and verifiable “train–evaluate–refine” transactional loop. Key innovations include an evidence-gated promotion mechanism, commit-key-based crash recovery, and a signed skill repository with a verified tool registry. The system enables zero-loss, zero-duplication execution and valid decision-making at arbitrary points under unattended operation, reducing erroneous promotions from 0.005–0.021 to 0.001 and successfully detecting all six classes of artifact tampering.
📝 Abstract
Coding agents such as Claude Code and Codex close the software loop: a main agent manages the loop, subagents analyze and execute, tools do the work. We port this architecture to robot-policy improvement, where one difference dominates the design: robotic tools---trained policies, training pipelines, data collection---fail routinely, so a tool's quality must be measured, recorded at every call, and expired when the artifact behind it changes. AgenticRobotics is a backend-independent control plane in which an LLM controller drives disposable workers through durable train--evaluate--improve transactions: an immutable objective, controller-owned measurement, commit-keyed crash recovery, an evidence-graded skill library, and a tool registry with a standardized, recorded call surface. The title is an operational claim, not a selection claim: the operator can leave because promotion is evidence-gated, state is recoverable, and capability quality is derived from records---not because the loop picks better checkpoints than a human; on the one lineage we measured, it does not. The gates measurably buy false-promotion control (0.001 per run hardened versus 0.005--0.021 shipped), anytime-valid decisions under optional stopping, zero lost or duplicate effects under kill injection, and six of six artifact-tampering classes caught by a signed verifier.