Xinduyuan Builds Robot 'Brain' with Uncertain Differential Geometry

In mid-July, inside Tsinghua Science Park, a humanoid robot, in an environment without the roar of GPU racks, faced an uncalibrated paper cup position. After a grasping command, it adjusted direction, walked to the table, steadied itself, grasped the cup, and lifted it—the entire process took over 10 seconds, uncut, with no hesitation or trial-and-error, at a human-normal speed.

The team behind this system is called Xindu Qiyuan. According to a report by Machine Intelligence, their algorithm did not use hundreds of GPUs for training, did not perform special calibration for the room, and did not even do a single pretraining for the cup-grasping action; from purchasing the robot body to deploying the algorithm, completing the full debugging from scratch took only one month.

This demonstration was used to contrast with the generalization boundaries and data bottleneck issues still debated by Silicon Valley embodied intelligence teams. Xindu Qiyuan claims its robots are using mathematics to counter fitting illusions, naturally completing the entire pipeline from voice command to grasping and delivery.

The technical foundation of Xindu Qiyuan is the uncertainty theory founded by Professor Liu Baoding of the Department of Mathematical Sciences at Tsinghua University. His representative work, "Uncertainty Theory," is widely cited and translated into Russian, Japanese, Chinese, and Persian, and is recognized by the international mathematical community as a branch of mathematics distinct from probability theory.

The theory distinguishes between randomness and uncertainty: phenomena satisfying the axioms of probability theory are called random, while those satisfying the axioms of uncertainty theory are called uncertain. Extensive empirical studies show that the real world is uncertain, not random; in reality, we observe neither the law of large numbers nor the central limit theorem, nor the law of the iterated logarithm.

The perfect self-consistency of probability theory rests on the implicit premise of frequency stability, but the real world lacks ideal conditions; home scenarios, factory disturbances, and weather do not simply repeat. When frequencies themselves are unstable, fitting with probabilistic models brings three defects: forced knowability, false precision, and data hunger.