| Popis: |
This benchmark presents an implementation of a standard control stack for an Autonomous Vehicle (AV). The control stack is made up of a behavioral planner (providing waypoints for the AV to visit in sequence), a trajectory planner (which computes smooth trajectory that the AV should follow to go between waypoints) and a trajectory tracker (which actuates the AV to make it follow the planned trajectory as closely as possible). The behavioral planner is purposefully simple, while the trajectory planner is a a high-fidelity approximation of a planner that was tested on a real Prius, and the tracker was validated by others in previous work on a real Cadillac SRX. The interest of this benchmark is that it includes all three components, rather than one AV control subsystem (such as only adaptive cruise control or only a lane-keeper), and the planners are significantly more realistic than most existing benchmarks or models. It can be used as a baseline AV system for verification and testing tools, which must be able to handle at least the complexity of this controller. This includes simple choices made by the behavioral planner when the current waypoint cannot be reached, discrete and continuous nonlinear optimizations solved by the trajectory planner, and nonlinear ODEs solved by the trajectory tracker. The bench- mark comes with three road topologies: a free space with obstacles, a curved road, and a roundabout. |
