Search papers, labs, and topics across Lattice.
6
0
12
End-to-end robot motion planning gets a boost: Flow Motion Policy leverages flow matching to generate diverse, collision-free paths, outperforming traditional methods in success rate and efficiency.
A surprisingly small 4B parameter model can achieve state-of-the-art multimodal embeddings by selectively applying chain-of-thought reasoning only when it demonstrably improves query-target alignment.
Ditch slow QR projections: Muon-accelerated updates slash computation time for tensor GLMs without sacrificing accuracy.
Contact-rich robotic manipulation gets a 14% performance boost by adaptively ignoring force/torque signals when they're irrelevant, and fusing them intelligently with vision when contact matters.
Forget hand-designed agent communication topologies: Agent Q-Mix learns decentralized communication strategies that boost accuracy and token efficiency in LLM multi-agent systems.
Achieve better compression in low-bit quantization by considering not just numerical sensitivity, but also the structural role of each layer.
