AI based tele-operation support through voxel based workspace modeling and automated 3D robot path determination

Abstract

The author is developing an intelligent remotely operable robot. Data transfer bandwidth limitations between the local master and the remote slave robot result in constraints on the quality and variety of perceptions that can be presented to the system operator. Machine intelligence at the robot end can alleviate this problem. Based on 'on-line' sensory data, the robot workspace occupancy can be modeled for effective and meaningful 3D representation to the robot operator. The sensory data abstraction is aimed at classifying the robot workspace into free or occupied space, path finding suitable for quick or precise approaches to 'parts to be handled', machine created advice for optimum routing, maneuverability, etc. It is important to develop methods for robust path creation for automated movement of the robot arm without hitting obstacles in the robot's 3D workspace volume. The paper describes the development of methods for 3D volumetric reconstruction of the workspace model through vision based sensing and 3D path finding in the presence of obstacles within the workspace for a 6DoF robot or gantry configuration. Techniques for creating realistic but light models of the workspace occupancy for the specific goal are described. Configuration space model development, obstacle free 3D robot path generation in the workspace and its remote visualization by an operator are dealt with in detail. Several optimization techniques and human intelligence integration are also described.