Sampling Based Predictive Control with Frequency Domain Input Sampling for Smooth Collision Avoidance*

This paper presents the smooth path generation method based on the sampling based model predictive control. Non-linear model predictive control framework is applied together with considering the car dynamics and non-linear constraints to achieve collision avoidance to obstacles. Since the control input must be optimized with non-linear constraints every control step in real-time, the problem is solved by sampling based approach. One of big issues on the sampling based approach is efficient sampling of smooth control input because some applications require the smoothness of control input. In proposed method, series of input are sampled in frequency domain directly in order to generate the smooth input series instead of applying the digital filtering to sampled input from specified random process. The proposed method samples the magnitudes of frequency components directly and those samples are transformed to time domain signals by Inverse Discrete Cosine Transform (IDCT.) Proposed method is implemented in simulation experiment with car dynamics simulator for collision avoidance task, and its validity is tested from the viewpoint of path tracking performance and real-time computation capability in several collision avoidance scenarios.