Mind-0

Mind-0 is an embodied robotics foundation AI model built by the Shenzhen-based Chinese startup MindOne Robotics (MindOn). It is a Vision-Language-Action (VLA) model designed as a single "mind" driving heterogeneous hardware platforms — from Unitree G1 humanoids to stationary dual-arm rigs. The core thesis of Mind-0 is that instead of training a separate model per platform on expensive teleoperation data, one can train one model on human-centric data (whole-body motion capture, egocentric cameras, handheld devices) and have it generalize across embodiments.

Two-layer architecture

Mind-0 decouples intelligence from embodiment. The high-level layer handles scene understanding, task reasoning, and behavior generation. The low-level Whole-Body Action Foundation Model — trained on tens of thousands of hours of motion-capture data — translates intentions into physical motion respecting each robot's dynamics, achieving sub-3 cm end-effector tracking accuracy while maintaining global motion coherence and balance.

Cross-Embodiment Data Pipeline

The cross-embodiment pipeline converts large-scale human demonstrations into action representations executable by different robots, effectively transferring human dexterity to hardware with fundamentally different kinematics, dynamics, and workspaces.

Real-World Execution Compensation Model

A lightweight compensation model trained on a small amount of real deployment data closes the sim-to-real gap. It corrects tracking errors, dynamics mismatch, and embodiment-specific deviations, reportedly achieving sub-1 cm manipulation accuracy on the Unitree G1 — a platform typically known for limited arm precision.

Hierarchical Coordination Reasoning

Human data is inherently delay-free, while robots suffer from perception and control latency. Mind-0 addresses this with a hierarchical reasoning loop — the high-level policy continuously monitors low-level feedback and adaptively decides when and how to invoke specific skills, rather than directly imitating human demonstrations.

Public demonstrations

Mind-0's first viral demo (November 2025) showed a Unitree G1 autonomously performing complex household chores with no speed-ups and no teleoperation. The second (June 18, 2026) showcased a heterogeneous fleet — two Unitree G1 humanoids and two stationary dual-arm rigs — running an end-to-end logistics workflow (shelf picking, transport, sorting, packing, tape sealing), with all four robots driven by a single Mind-0 model.