LATIDIA · Robótica

Menos tokens, mejor acción: agentes robóticos Astra GPT-6 con una tasa de éxito un 14% más alta pero un 65% menos de tokens

arXiv: 2610.01939v1Tipo de anuncio: Cross Resumen: Los agentes del modelo de lenguaje de visión (VLM) pueden controlar los robots a través de la retroalimentación visual y las primitivas de acción, pero las invocaciones repetidas del modelo y las observaciones redundantes incurren en sub

La noticia

arXiv:2610.01939v1 Announce Type: cross Abstract: Vision language model (VLM) agents can control robots through visual feedback and action primitives, but repeated model invocations and redundant observations incur substantial token overhead. We introduce PyRUA-Lean, an interactive code-execution framework that couples feedback-driven primitive composition with selective observation: the agent composes classical robot primitives and learned vision-language-action (VLA) policies into Python cells that perform conditional checks and local retries, returning only explicitly requested images and state feedback for replanning. Across 700 simulated task instances from LIBERO-PRO, RoboTwin 2.0, and RoboCasa365, we compare PyRUA-Lean with a tool-calling baseline using the same GPT-6 Astra planner and underlying robot primitives. Under equal LLM-call budgets, PyRUA-Lean increases