LATIDIA · Robótica

CAPEX: Destilación eficiente del comportamiento del modelo de base en políticas de robots desplegables a través del razonamiento adaptativo de la experiencia

arXiv: 2609.33007v1Tipo de anuncio: nuevo Resumen: El aprendizaje de robots se ha basado en gran medida en demostraciones operadas por humanos para adquirir comportamientos aprendibles efectivos. Sin embargo, los procesos de recopilación de datos operados por humanos pueden ser

La noticia

arXiv:2609.33007v1 Announce Type: new Abstract: Robot learning has largely relied on human-teleoperated demonstrations to acquire effective learnable behaviors. However, human-operated data collection processes can be unintuitive, difficult to scale, and inherently asynchronous. We explore an alternative: distilling physical behavior from general-purpose multimodal foundation models into deployable robot policies by using the foundation model itself as an autonomous demonstrator. While sufficiently capable models can generate successful zero-shot manipulation trajectories, repeatedly invoking them during physical execution is slow and expensive, limiting their utility as scalable data generators. As a solution, we introduce CAPEX, an experience-conditioned demonstration collection framework that uses execution experience from previous attempts to adapt how frequently the foundation model