Realtime collision avoidance control based on continuation method for nonlinear model predictive control with safety constraint

Both of a path planning and a controller to follow the planned path are indispensable functions to realize an autonomous driving. While a controller to track the desired path obviously requires the plant model of the car to determine its input, it is necessary to consider the car dynamics and the physical limitations in path planning since the feasibility of the path is depending on them. From this view point of that the design process of those functions are inseparable from each other, simultaneous motion planning and control (SMPC) is attracting a great attention as the framework which can deal with those two functions in single optimization problem. This paper approaches the automated collision avoidance based on SMPC framework utilizing the nonlinear model predictive control (NMPC). Instead of planning the detailed path for the controller, simple references and/or constraints are given to the NMPC problem to realize the collision avoidance and lane keeping. In order to realize a real-time optimization of NMPC problem, Continuation and GMRES method (C/GMRES method) which can archive an extremely fast computation for NMPC is introduced. Two ways to realize the collision avoidance, which are the realization by the constraint not to enter the collision area and by changing the reference to avoid the obstacle, are implemented and tested to compare those performances to each other.