Computationally Efficient Dynamic Window Approach Based on Pattern Search Optimization

Collision avoidance is a crucial aspect of autonomous ground vehicles (AGVs). One of the most common algorithms is called the dynamic window approach (DWA). The algorithm enables AGVs to operate with high performance in an unknown environment with a particular emphasis on achieving maximum linear and angular acceleration. However, DWA requires high computational effort to examine all possibilities with high resolution, and then select the best possible pair of control signals, i.e., linear and angular velocities. In this paper, the Pattern Search (PS) optimization algorithm is used to reduce the computational requirement of the DWA. Instead of calculating the DWA objective function for each possibility, the PS is used for algorithmically selecting the next examined pair of control signals. The results obtained demonstrate that a similar resolution of control signals can be achieved with almost two times less computational effort. The proposed approach has been examined in the MATLAB environment, while the source code is available on the MathWorks FileExchange.