AFM – A Universal Test Vehicle for a Wide Variety of AI-based Applications

The AI For Mobility (AFM) concept has been designed as a universal platform for testing of vehicle dynamics control methods. A classification of relevant application fields can be illustrated using a pyramidal control architecture. It comprises three levels related to the corresponding fields of application as follows.

Vehicle Application

The top level deals with the vehicle application and is used to generate trajectories or paths. Reactive trajectory optimization is performed on the basis of environmental information. For this purpose, the AFM is equipped with various perception sensors such as cameras, lidar or radar sensors. By means of advanced machine learning methods, the driving environment is recorded and processed in real time. The resulting motion requirements for the three planar degrees of freedom are passed to the Motion Execution.

Motion Execution

This level contains the vehicle dynamics control, which calculates appropriate vehicle dynamics quantities to realize a required motion and ensures safe operation of the vehicle. The AFM’s extensive sensor equipment enables the development and testing of advanced AI control approaches. The horizontal vehicle motion of the AFM described by the longitudinal (throttle and brake) and lateral (steering) dynamics is operated by a full by-wire kit. In order to control the vehicle’s vertical dynamics, the AFM is equipped with semi-active dampers. The resulting control outputs, i.e. actuator demands, are then considered at the Chassis Level.

Chassis Level

The lowest level deals, for instance, with the smart actuator control of the steering and braking system. In the AFM, extensive information about the actuators is available for many sub-systems (e.g. by reading the vehicle CAN buses). Due to this feature, the AFM provides a suitable test platform for applications in this level.

Some applications such as state estimation methods may span over all three levels.

The holistic capability for testing novel AI-based control methods makes the AFM a unique test platform. Most other vehicle demonstrators do not take all three levels into account leading to a research gap in the context of vehicle dynamics.

This article is partly based on the open access publication “AI-For-Mobility – A New Research Platform for AI-Based Control Methods“.