LATIDIA · Robótica

Aprendizaje del comportamiento de los robots a partir de datos sintéticos no robóticos

arXiv: 2610.31275v1Tipo de anuncio: nuevo Resumen: Los datos reales de robots, ricos en tareas y morfologías, a menudo se tratan como el cuello de botella para entrenar modelos de aprendizaje de robots, incluidos los modelos de cimientos incorporados que apuntan a

La noticia

arXiv:2610.31275v1 Announce Type: new Abstract: Real robot data, rich across tasks and morphologies, is often treated as the bottleneck for training robot learning models, including embodied foundation models that aim for zero-shot physical transfer. In this paper, we investigate the degree to which such models can acquire necessary structure using data from systems without embodiment. To this end, we test if inexpensive, procedurally generated synthetic data can yield useful priors before a model observes any real-world behavior. We study this through robot action prediction, a natural pretraining objective and a direct test of whether structure learned from arbitrary synthetic tasks transfers to embodied settings. We introduce three data generation pipelines