1 paper
Marcus Fechner, Hamza Adnan, Constantin C. Lüth +3
Latent action models (LAMs) offer a promising path to pre-training embodied agents on large amounts of action-free video. They infer latent actions between consecutive observations…