Galaxy General chooses simulation synthesis method to collect humanoid robot data

Galaxy General chose the method of simulation synthesis to collect data for humanoid robots. They developed and synthesized tens of millions of scene data and billions of grasping and navigation data, and proposed DexGraspNet, the world's largest dexterous hand dataset. Although the data scale is positively correlated with the final performance of the model, the speed of performance improvement is not proportional to the growth of scale.