Zhiyuan announced on October 28 an upgrade to its one-stop embodied-intelligence simulation platform, Genie Sim, breaking through the bottleneck of flexible-object simulation, a Daily AI Morning report said.
Built on Real2Sim real-world system identification, a self-developed physics engine and world-model capabilities, the platform achieved a 96.7% correlation between simulation and real-robot task performance across zero-shot tests covering four robot bodies and 22 complex tasks.
The upgrade links scene reconstruction, data generation, automatic evaluation, post-training and open-world exploration, letting simulation results serve directly as a trusted reference for real-robot outcomes - a key link in moving embodied robots from algorithm training to deployment.
Flexible objects such as clothing, cables and packaging have long been blind spots for simulators; closing that gap is expected to lower the real-machine trial-and-error cost that dominates home-service and light-industrial deployment.
Source: Daily AI Morning (paraphrased and translated from Chinese media; reprinted for informational purposes with attribution).
