Intelligent Robot launches ACoT-VLA and open sources it.

On March 9, Zhiyuan Robotics announced that the ACoT-VLA architecture jointly proposed by the company and Beihang University was officially selected for the Computer Vision Top Conference CVPR 2026. This research broke through the bottleneck of the traditional VLA model with "strong semantics, weak actions" and proposed for the first time a thinking chain paradigm for inference in the "action space". Currently, ACoT-VLA has been officially open-sourced as the official baseline model for the AGIBOT WORLD CHALLENGE, helping developers worldwide to overcome the challenges of complex robot control.