Nonlinear Model Predictive Planning and Control for High-Speed Autonomous Vehicles on 3D Terrains

A novel model predictive formulation for autonomous vehicles to plan and execute collision-free and dynamically feasible maneuvers on 3D terrains is introduced. Common approaches for navigating on 3D terrain often rely on graph search techniques or other simplified 2D models to predict the plant behavior. On 3D terrains, it is hard to take vehicle dynamics into account efficiently during planning and control. To address this gap, a vehicle model that considers terrain topology is constructed as the prediction model. A single layer nonlinear model predictive control framework is used to optimize the control inputs of steering rate and longitudinal acceleration based on the newly introduced vehicle model. The new framework is evaluated in simulation on a High Mobility Multipurpose Wheeled Vehicle (HMMWV) climbing on a terrain with varying slopes. Results show that the conventional methods produce failing maneuvers, whereas the new algorithm successfully navigates the vehicle to the target. Copyright (C) 2021 The Authors.

Automated summary

A novel model predictive formulation is introduced for autonomous vehicles to plan and execute collision-free and dynamically feasible maneuvers on 3D terrains. By constructing a vehicle model that considers terrain topology and using a single layer nonlinear model predictive control framework, the control inputs are optimized for steering rate and longitudinal acceleration. The new algorithm successfully navigates the vehicle to the target on varying slopes terrain, outperforming conventional methods in simulation.