Robots performing tasks such as tightening screws, hammering and remote assembly are part of a system announced by Beijing Galbot, which has integrated these capabilities into a teleoperation framework powered by its new dexterous manipulation model, DexNDM. Developed with Tsinghua University, Peking University, the University of Adelaide and Zhejiang University, the model is described by the company as a neural dynamics system designed to manage in-hand object rotation across a wide range of shapes and sizes.
According to Galbot, DexNDM adjusts to changing forces and varying wrist orientations during multi-axis movements. The firm states that operators provide general instructions while the model controls detailed finger motions, enabling robots to handle tasks requiring fine manipulation. Galbot also outlines a training method built on task-specific policies that are combined into a single model intended to work across different robot types and environments.
The company also introduced NavFoM, a navigation model described as applicable to various robot forms and navigation tasks. Galbot reports that NavFoM interprets video input and text commands to generate movement paths and can operate indoors and outdoors without advance mapping. The model is presented as capable of target-following, autonomous navigation and responding to simple language instructions, with applicability to quadrupeds, wheeled platforms, drones and vehicles. Galbot positions both models as tools for deployment in industrial, logistics and other settings requiring autonomous movement or fine-grained manipulation.
