IR Vision and EKF Filtering
The tracking system is based on infrared (IR) vision and extended Kalman filtering (EKF). The track is illuminated by IR light. The reflective markers on the cars reflects the light into a camera with a longpass IR filter, mounted above the track. The reflected pattern is unique for each car, and is detected in every camera frame. The noisy sequence of car positions obtained from the camera sometimes contains false and missed car detections, and is filtered using EKF. By this, we can estimate the position with very good precision. Thanks to the model of the dynamics in the EKF, we can also make sensible estimates even when the car is hidden for the camera for a few seconds. The filtered position is then fed into the controller.
Particle Filter – an alternative
As an alternative to EKF, particle filters have been investigated. The EKF relies on a linearized model of the car dynamics. The particle filter enables the use of a highly non-linear model with, e.g., a discrete slip/grip parameter. The downside is the computational load of the particle filter. We have managed to achieve sensible tracking results with the particle filter, but it is not used in the current version of the system.
Offline optimal trajectory
The control action is divided into two layers. The first layer is an (optimal) trajectory computed offline, and the second layer is an online controller (acting in real-time) working with the pre-determined trajectory as a reference value for the speed and position.
The optimal trajectory was computed offline for a car model (the Single Track Model, see the project reports for details) using the modeling tool Modelica and optimal control theory. The optimal trajectory contains both a speed and a position reference. The position reference of the optimal trajectory is:
Online PID control
In realtime, the cars are controlled to follow the reference trajectory. The control is done with classical PID control. The control action is computed in the workstation, and sent to the cars using the radio communication and a modified hand controller.
Another alternative to the controller, investigated at some other similar race tracks, is MPC. With MPC, a more advanced control action can possibly be taken, e.g., utilizing slip.