Overview
MiniCPM-Robot is an embodied intelligence model series jointly open-sourced by Mianbi Intelligence and OpenBMB, including the general VLA model MiniCPM-RobotManip (1.5B parameters) and the mobile tracking model MiniCPM-RobotTrack (0.9B parameters), focusing on robot manipulation and tracking capabilities on edge devices.
Key Features
- VLA General Manipulation: RobotManip 1.5B parameters, a vision-language-action model supporting general robot manipulation
- Mobile Tracking: RobotTrack 0.9B parameters, focusing on mobile tracking capabilities
- Fully Open Source: Model weights and code are open, enabling local deployment
- Edge-Friendly: Small parameter count suitable for real-time operation on edge devices
Use Cases
- Robot manipulation control
- Mobile platform tracking
- Embodied intelligence research
- Edge device AI deployment
Pros
- Small parameters suitable for edge deployment
- Fully open source and customizable
- Dual model coverage for manipulation and tracking
Pricing
Open source and free, based on MIT or similar license, subject to the repository.
Summary
MiniCPM-Robot extends Mianbi's edge-side small model capabilities to embodied intelligence. The combination of 1.5B manipulation and 0.9B tracking models is suitable for robot scenarios on edge devices, and it is open source and customizable.
Version History
- MiniCPM-Robot series open source release (2026-07-20): Includes 1.5B VLA manipulation model and 0.9B tracking model, released and open-sourced by Mianbi Intelligence and OpenBMB