Modelbest Open-Sources Embodied AI Model MiniCPM-Robot
AI寒武纪 · wechat · 2026-07-19
Modelbest announced and open-sourced its first embodied intelligence model series, MiniCPM-Robot, at WAIC, featuring two main branches:
- MiniCPM-RobotManip: A general VLA model for robotic manipulation, focusing on solving the 'memory' issue in long-horizon tasks. The article explains how it utilizes visual Token compression and streaming inference to reduce the compute cost of incorporating historical observations to near that of single-frame reactive models. It ranks among the top tier on mainstream VLA benchmarks and significantly outperforms traditional methods on RMBench.
- MiniCPM-RobotTrack: A local tracking and navigation model for robot dogs, supporting single-target, dynamic multi-target, and fuzzy command tracking. It emphasizes fully local, offline usability, with official success rates topping other open-source solutions.
The article also introduces the supporting inference framework PhyAI, alongside real-world partnerships with Unitree Go2Edu, AUBO Robotics, and Geely Automobile, covering applications like showroom guidance, park patrolling, and long-horizon warehousing tasks. The overarching theme is Modelbest's successful transfer of its expertise in edge-side multimodal compression and inference optimization into the embodied AI domain.
Related event: ModelBest Open-Sources Embodied Model Series MiniCPM-Robot at WAIC(6 posts)→
More from Embodied
- Amazon and Google sold 600M+ smart speakers, so why no AGI-era successor? — julianlehr · 2026-09-11
- ECCV26 Oral: Flow Matching Enables Single-Stage Multi-View Point Cloud Registration — ducha_aiki · 2026-09-11
- Polish developers build iPhone app that detects nearby Meta smart glasses — Low-Honeydew6483 · 2026-09-11
- Ant's Afu health AI hits 150M users, unveils AI+hardware health alliance at Bund Summit — APPSO · 2026-09-11
- Johns Hopkins Launches Full-Stack Hands-on Robot Learning Class with SO-101 Arm Kits — _krishna_murthy · 2026-09-11
- SyncWorld: In-Context Robot World Model Simulates Unseen Views and Embodiments Zero-Shot — ChongZzZhang · 2026-09-11