In an October 28 announcement, AGIBOT described a major Genie Sim update connecting scene reconstruction, flexible-interaction simulation, capability evaluation and policy post-training.
The company says it evaluated real-world-trained models in simulation without adapting the simulated environments. Across four robot platforms and 22 tasks, it reports 96.7% correlation between simulated and real-robot task scores.
AGIBOT says the upgraded benchmark adds dexterous manipulation, whole-body coordination, fine manipulation and transfer across robot embodiments. It also describes a Genie Sim × RLinf parallel post-training framework that uses evaluation results to select tasks for reinforcement learning.
Companies in this report
Supplier profiles in the RoboPallas directory