LTV-MPC Based Trajectory Planning Considering Uncertain Object Prediction Through Adaptive Potential Fields
Florian Siebenrock, Manuel Gunther, Sören Hohmann · 2020 IEEE Conference on Control Technology and Applications (CCTA) · 2020
In this paper a combined longitudinal and lateral trajectory planning approach is presented using an linear-time-varying model predictive control (LTV-MPC) scheme. To consider uncertain moving object state and prediction information, an adaptive potential field is modelled generating a time-varying obstacle cost-term. Approximating the nonlinear planning problem along the previous planned control and state sequence results in an LTV-MPC planning problem and ensures small approximation errors of the original nonlinear formulation. The proposed LTV-MPC planning approach is analyzed in a collision avoidance scenario with a pedestrian. The results show a more defensive maneuver integrating the uncertain object prediction information in the planning problem and the proposed approximation scheme delivers a difference of 492,18 % in the relative approximation error of the cost function in the presented scenario compared to the widely used approach, approximating the planning problem by using the current vehicle state.