Technology

Brain-Inspired System

Active memory, flexible judgement and self-improvement through coordinated minds

A separated dual model pairs a cerebrum for perception and decision-making with a cerebellum for motion control, replacing the single end-to-end VLA approach that dominates the field (OpenVLA, Pi and similar).

Architecture: cerebrum and cerebellum

Architecture: cerebrum and cerebellumText promptScene imageMultimodal feature extractionCerebrumPrior state and task plan sequenceCerebellumNoised actionDiffusion /timestepGenerated action"Open this package for me"Cross-attentionCurrent instruction and stateUpdated task sequence
  • Cerebrum

    A multimodal large model for state perception and high-level task decisions

  • Cerebellum

    Motion generation built on a diffusion model

How this differs

Conventional embodied AI: command – action – feedback, point to point

Brain-inspired system: task analysis added at the algorithm layer