Fast Robot-Obstacles Distance Evaluation in the Depth Space

A robot-obstacle distance evaluation algorithm is described in this thesis which directly operates on the image given by a depth sensor, also known as Depth Space. The algorithm’s main features are that the entire visible part of the robot is accurately surveyed and that the different robot links are separately considered for the minimum robot-obstacle distance. Several design parameters are described and evaluated, introduced mainly to improve the algorithm’s time performance. Possible implementations of the algorithm with the respective design parameters are given both for a CPU and a GPU. Performance results of the CPU and GPU implementations, in terms of accuracy and computational time, are provided and compared for different parameter values. The implementation on the GPU outperforms the one on the CPU clearly, especially concerning the execution time. With well chosen parameter values, the algorithm on the GPU can reach cycle times of 1000 Hz, by simultaneous providing accurate results.