Vehicle Dynamics Models for Onboard Motion Planning
Offline multibody dynamics based modeling and simulation of vehicle dynamics has been pursued with varying levels of success for more than two decades. This has been used in design, controls, training, and other technical and programmatic objectives. Over the last decade, autonomous vehicle dynamics has become an important area of research. This has resulted in a growing need for onboard vehicle model that works with the vehicle controller and path planner. Typically, kinematic models have largely been used for these objectives. Use of dynamics models for onboard motion planning is a relatively new topic of research with only a handful of prior work. In this paper we report our attempts at addressing the need for onboard vehicle dynamics models for motion planning in relatively fast autonomous mobility scenarios. We present the idea of using adaptive motion models that trade fidelity and cost of simulation to enable a motion planner to select an adequate model. Towards this, we present representative simulation results that demonstrate the need for adaptivity. We then present some technical challenges with onboard vehicle models and our attempts at addressing these challenges. Finally, we present some results that compare raw vehicle data with model predictive results.