EVENT DOSSIER
On September 8, 2026, DaXiao Robotics, in collaboration with S-Lab at Nanyang Technological University and Shanghai Artificial Intelligence Laboratory, officially launched HSImul3R, the world’s first framework for simulating human-scenario interaction reconstruction. This achievement aims to bridge the “perception-to simulation gap” between 3D vision and physical simulation. It introduces a physical closed-loop optimization mechanism, incorporating gravity stability and real-world contact into core evaluation criteria, and establishes the HSIBench verification system. Experimental data shows that HSImul3R achieves interaction stability rates of 53.68%, 30.56%, and 13.92% in easy, medium, and difficult tasks, respectively, significantly outperforming traditional methods; the human-scenario 3D mapping rate decreases from 69.51% to 22.90%. The team has successfully applied the optimized algorithm to the Unitree G1 humanoid robot, achieving a complete link from human video input to real-world robot execution. This achievement has been officially accepted by ECCV 2026.
近日，大晓机器人联合南洋理工大学 S-Lab、上海人工智能实验室发布全球首个可仿真人—场景交互重建框架 HSImul3R。该成果针对三维视觉与物理仿真间的“感知—仿真鸿沟”，提出物理闭环双向优化框架，将重力稳定性、真实接触等纳入核心评价标…
2 reports大晓机器人联合南洋理工大学 S-Lab、上海人工智能实验室发布全球首个可仿真人–场景交互重建框架 HSImul3R。该成果针对三维视觉“感知—仿真鸿沟”，提出物理闭环双向优化机制，将重力稳定性、真实接触纳入核心评价标准，并构建 HSIBench 验证体系。实验显示，HSImul3R 在易、中、难任务中的交互稳定率分别为 53.68%、30.56% 和 13.92%，显著高于传统方法；人–场景三维穿模率从 69.51% 降至 22.90%。团队已将优化后动作迁移至 Unitree G1 人形机器人，贯通人类视频到真实机器人执行的完整链路，推动视觉信息转化为可仿真、可学习的机器人技能资产。
近日，大晓机器人联合南洋理工大学 S-Lab、上海人工智能实验室发布全球首个可仿真人—场景交互重建框架 HSImul3R。该成果针对三维视觉与物理仿真间的“感知—仿真鸿沟”，提出物理闭环双向优化框架，将重力稳定性、真实接触等纳入核心评价标准，并构建 HSIBench 进行检验。HSImul3R 在易、中、难三档任务中的交互稳定率分别为 53.68%、30.56% 和 13.92%，显著高于传统方法，并将人—场景三维穿模率从 69.51% 降至 22.90%。团队已将优化后的人体动作迁移至 Unitree G1 人形机器人，贯通视频输入到真实执行的完整链路。该成果已被 ECCV 2026 正式接收。