Release · Sep 13, 2026
Liangyuan Xinchuang releases LightParkour, LightNav-0 and Light REACT embodied AI technologies
Three embodied AI technologies have come from Liangyuan Xinchuang in just over a month. LightParkour uses physical simulation and curriculum learning to expand complex contact skills and distills walking and parkour abilities into one unified policy. LightNav-0’s Real2Sim2Real data engine converts more than 2,000 real scenes into simulation and creates over 4,000 hours of vision-language-action experience for post-training general navigation. Light REACT handles whole-body resilience under external disturbances, falls, and hardware damage. The company places the releases in a Physical AI path of scaled pre-training, alignment, and deployment. Commercial availability was not specified.
Original sources (Chinese)
2000+真实场景搬进仿真!一个导航模型零样本“通吃”四种机器人本体