Search papers, labs, and topics across Lattice.
Affiliation:
3
0
3
Multi-agent path finding can reliably scale to 100,000 agents in dense 3D spaces by treating neural networks as counterfactually gated advisors over classical priority inheritance rather than end-to-end controllers.
CometVLA reveals that integrating embodied physical commonsense into VLA models leads to substantial improvements in real-world manipulation tasks, challenging the notion that disembodied training is sufficient.
Achieving zero floating and unstable bricks, ResemBrick redefines how we approach 3D reconstruction by tightly coupling visual fidelity with structural integrity.
