Improvement of Control Performance of Sampling Based Model Predictive Control using GPU

Real-Time Implementation of Randomized Model Predictive Control for Autonomous Driving

“…Based on these considerations, this paper addresses the RMPC as a solution for autonomous driving control. First, it extends the previous works [17] and [30] with an improved vehicle model, thereby removing the limitations of driving only in straight line and with a constant speed. Then, it solves the problem of randomness by frequency domain sampling and implements the RMPC along with the new sampling methodology.…”

“…In addition, the ego car being represented with a bicycle model, this study concluded that more computational power is necessary to use the RMPC in real-time. This demand for computational improvement was also investigated in our previous works [29] and [30]. Computation speed of linear MPC problems was accelerated using GPU in [29], however, this was limited to linear MPC problems.…”

“…Computation speed of linear MPC problems was accelerated using GPU in [29], however, this was limited to linear MPC problems. On the other hand, [30] discussed some preliminary results showing the GPU's potential to accelerate RMPC. This study was also limited to simulations and to straight line driving at constant speed.…”