Visual Perception for Autonomous Racing Car Competition
EUFS is a student group at the University of Edinburgh which competes at the Formula Student Challenge, where each team builds from the ground up its own Racing Car, one of the categories is for Autonomous Vehicles, where each team competes to finish a series of laps around a previously unseen track autonomously. Inside the Artificial Intelligence team, I am part of the perception team. We are in charge of building the perception stack in order to pass all the relevant semantic information regarding the track to planning. The Perception team works with a set of sensors, mainly Lidar and Stereo Cameras. I work with the Stereo Cameras to abstract the information of the limits of the track and position of the car (basically state estimation). The track is a closed loop limited by traffic cones at the sides, so I need to recognize the position, type and depth on the plane and merge it with the information coming from the Lidar. Once we abstract this information it is time for the team in charge of planning.
The main architecture is based on sensor fusion coming from redundant information from Lidar and Stereo Cameras. More exactly cloud points in the ground plane as well as semantic information from cameras. The system implemented applies a filter to combine both sources of information, which we are working hard to improve to an Error State Extended Kalman Filter.
The first image is a picture of the Race Car developed by the team. In the following year, the objective is to add the necessary actuators and the hardware stack to make it fully autonomous for this task. Meanwhile, for the competition, all teams are provided with a self-driving race car (see image below).
Finally, the last image is the frame of an old version of our car next to a demo of the stack during this year presentation of our progress to the university and to the sponsors, you can see the parts of the pipeline, in this case, raw image, detection, lidar and path optimization.
* Images taken from EUFS website.