Model Simplification for Efficient Collision Detection in Robotics

Motion planning for industrial robots is a computationally intensive task due to the massive number of potential motions between any two configurations. Calculating all possibilities is generally not feasible. Instead, many motion planners sample a sub-set of the available space until a viable solution is found. Simplifying models to improve collision detection performance, a significant component of motion planning, results in faster and more capable motion planners. Several approaches for simplifying models to improve collision detection performance have been presented in the literature. However, many of them are sub-optimal for an industrial robotics application due to input model limitations, accuracy sacrifices, or the probability of increasing false negatives during collision queries. This thesis focuses on the development of model simplification approaches optimised for industrial robotics applications. Firstly, a new simplification approach, the Bounding Sphere Simplification (BSS), is presented that converts triangle-mesh inputs to a collection of spheres for efficient collision and distance queries. Additionally, BSS removes small features and generates an output model less prone to false negatives.