Li Kang

MA-VLA achieves robust multi-arm compositional generalization, outperforming existing models by effectively enabling role-agnostic behavior in unseen collaboration scenarios.

Generating robot training data that bridges the sim2real gap doesn't require painstakingly detailed simulation environments; instead, a neural simulator can transform classical simulations into realistic representations using only a small amount of real-world data.

Stop hard-coding reasoning strategies for your LLM agent: a learned router that dynamically picks the best paradigm for each task boosts performance by up to 5.5%, beating even the best fixed strategy.

Coordinating embodied multi-agent systems doesn't require end-to-end training; instead, offload planning to a VLM in simulation and transfer back to the real world for execution.