MiniCPM-RobotManip is a 1.5B VLA that claims half the vision compute

iamfakhrealam · x · 2026-07-24

MiniCPM-RobotManip is a 1.5B general-purpose VLA that is positioned against Qwen-VLA and π0.5.

The post says it uses roughly half the visual-processing compute of those systems, with the main trick being efficient visual token compression. The accompanying chart frames the compute savings as history length grows.

Related event: OpenBMB Open-Sources MiniCPM-Robot for Offline Embodied AI(10 posts)→

Original post →

More from Embodied

Embodied channel →