The project aims to develop new methods for representing the environment used in robotics that will exploit recent advances in neural and parallelized 3D data processing. The result will be a fast and accurate model of the environment that can be used by neural motion planning methods that ensure the computational efficiency and accuracy of motion planning algorithms. The project will also develop methods for neural representation of robot motion constraints. We plan to focus on the various network architectures that can be used to take into account motion constraints like self-collisions, workspace, joint load, etc. The proposed models will be explainable, physically interpretable, and differentiable, which will allow rapid avoidance of motion constraints during motion planning. The project also aims to demonstrate the effectiveness of such algorithms in typical tasks performed by manipulating and mobile robots.

The work was supported by the National Science Centre, Poland, under research project no UMO-2023/51/B/ST6/01646.