88Z AI, 21 July — ModelBest released the open-source embodied AI model series MiniCPM-Robot at WAIC 2026, featuring the general manipulation model RobotManip and the tracking navigation model RobotTrack. The former achieves high performance with small parameters and supports native memory, while the latter is the first tracking model supporting local offline deployment.
At a glance
- RobotManip achieves high performance with small parameters, supports native memory, and halves inference costs
- RobotTrack is the industry's first tracking model supporting stable local offline operation
The update
During the World Artificial Intelligence Conference 2026, ModelBest officially open-sourced its first embodied AI research outcome, the MiniCPM-Robot model series. This series mainly includes two core components: the general Vision-Language-Action model RobotManip and the tracking navigation model RobotTrack. RobotManip achieves performance comparable to larger models with minimal parameters, supporting one minute of native memory while halving inference costs. Meanwhile, RobotTrack is the industry's first tracking model that supports local offline deployment, ensuring stable operation in environments without internet connectivity, thus providing significant support for offline applications of embodied AI devices.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 面壁智能Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.