LaLa-Policy
A policy model initialized from a language-conditioned latent world model that jointly predicts actions and future visual representations; on the LIBERO and RoboCasa benchmarks its performance is comparable to Cosmos Policy, with faster inference.
Timeline (last 90 days)
A pretrained LaLa-World can serve as initialization for LaLa-Policy, a policy model that jointly predicts actions and future visual representations.
On the two benchmarks LIBERO and RoboCasa, LaLa-Policy performs comparably to Cosmos Policy, with inference up to 5.5 times faster.
All Facts (2)
A pretrained LaLa-World can serve as initialization for LaLa-Policy, a policy model that jointly predicts actions and future visual representations.
50%UnverifiedOn the two benchmarks LIBERO and RoboCasa, LaLa-Policy performs comparably to Cosmos Policy, with inference up to 5.5 times faster.
50%