In robotics, movements must be precise and well-controlled. As robots are asked to do more as industry becomes more familiar with the technology, there will be an increasing need for high accuracy ...

(a) The main structure of the self-supervised pretraining model, including three parts—a token embedding at the forefront, followed by a hierarchical encoder–decoder and a point reconstruction module.