Guangxiang Technology, a startup spun off from Tsinghua University's laboratory founded just one year ago, has officially launched its industrial-grade self-evolving embodied intelligence robot, the Phi-Bot X1. Instead of chasing humanoid form factors, the company has taken a pragmatic 'subtraction' approach to design, directly targeting real pain points in automotive manufacturing.
Design Philosophy: Subtraction Over Showmanship
Phi-Bot X1 features a four-wheel omnidirectional steering chassis instead of bipedal legs — more stable and maneuverable in narrow factory aisles, capable of active steering, lateral crab-walking, and in-place rotation. It uses two-finger grippers instead of dexterous hands, following Occam's Razor: in industrial settings, complexity does not equal superiority. The robot achieves 27 degrees of freedom with a 0-2.5m industrial lifting waist, 1kHz full-joint force-controlled双臂, and can work while moving.
Key Specifications and Capabilities
The robot is equipped with a multi-dimensional perception system including 3D LiDAR, RGBD depth cameras, binocular cameras, and ultrasonic radar, achieving 10mm positioning accuracy and 0.05mm end-effector repeatability. Through replaceable end effectors, a single Phi-Bot X1 can cover quality inspection, material loading, sorting, tightening, adhesive application, and assembly tasks.
Deployment time is dramatically reduced — from 6 months for traditional automation to just weeks or even days — because the robot relies on its on-board perception rather than requiring massive factory line retrofitting.
Real Factory Validation
Phi-Bot X1 has already entered real production line collaboration with multiple leading automakers. In body surface inspection tasks, it achieves 25-45% efficiency improvement over manual labor and 51% improvement over non-collaborative automation. In welding feeding stations, it completes long-horizon complex tasks including grasping, moving, flipping, precision alignment (millimeter-level dynamic positioning, angle control within 0.3°), and placement.
At the 2026 ATC exhibition, Phi-Bot X1 demonstrated 21.5 hours of continuous loading/unloading operation with zero errors and zero interruptions over three days.
Self-Evolution via Data Flywheel
Powered by a tripartite technical system — Phi-RL Matrix reinforcement learning algorithms, Phi-Space efficient data pipeline, and Phi-Arch physical intelligence development platform — the robot supports progressive self-improvement as deployment cases accumulate. Guangxiang's strategy focuses on automotive manufacturing as the entry point, with plans to expand to broader industrial scenarios.
