LDA-1B Foundation Model Trains On 30,000 Hours Of Human And Robot Data
Reactions from ranked influencers
2 posts
LDA-1B is a unified world model - jointly trained on 30,000 hours of human and robot data, to predict actions, forward-simulate the world in a latent representation, and predict possible futures. It's a good look at what the future of robot foundation models might look like ->
Instead of choosing between training a world model and training a language conditioned robot policy, why not do both? LDA-1B s a new foundation model that is trained on 30,000 hours of human and robot interaction data. Part of the secret is that LDA-1B jointly learns forward…
One of the more interesting innovations here is the use of a meaningful, structured DINO latent space, which captures spatial structure over merely predicting pixels