Search papers, labs, and topics across Lattice.
6
0
5
7
KineBench reveals that embodied world models exhibit task-complexity-bounded nonlinear scaling, challenging existing assumptions about model performance and data requirements.
UMI-Bench 1.0 reveals that standardized real-world evaluations can dramatically improve the reliability of UMI-style robotic manipulation policies.
SpaceVLN achieves state-of-the-art zero-shot navigation performance by integrating spatial cognitive memory with task-guided reasoning, redefining how agents understand and navigate complex environments.
OASIS shows that simulation can outperform real-world teleoperation in humanoid manipulation tasks by leveraging diverse environmental variations.
By pretraining a VLA model with goal-conditioned RL, PRTS learns to reason about goal reachability, leading to substantial gains in long-horizon robotic tasks and zero-shot generalization.
LLMs can generate physically plausible and semantically consistent motions from text descriptions previously unseen in training data, if you guide their reasoning with MCTS and refine the results with physics-aware RL.
