UBTECH released its Thinker-Sim embodied-intelligence simulation platform on November 2, built around a Real2Sim2Real core to address the high cost of real-data collection, the complexity of simulation-environment construction and long algorithm-validation cycles.
The platform offers an end-to-end system covering real-scene digitization, scaled simulation-data synthesis and robot-capability evaluation. Its self-developed Thinker-GS high-fidelity scene-reconstruction algorithm improves rendering efficiency by 67.3%, while a single node (8 GPUs) can synthesize 390 hours of simulation data per day, with communication efficiency up more than 400% and a visual workflow fitting over 90% of target scenarios.
The self-developed Ulab-Agent supports natural-language task orchestration and scheduling, lowering the barrier to complex simulation. The release strengthens China's embodied-AI toolchain as humanoid robots move from demos toward real deployment.
Source: China News Report (compiled from Daily AI Morning).
