Comparisons of Platooning Control Methods without Inter-Vehicle Communication

Abstract

Platooning is when one or more follower vehicles follow a lead vehicle autonomously based on information gained through sensors or communication. Due to being autonomous, the vehicles in the platoon can react faster and drive closer to each other than usual. The faster reaction increases safety. The smaller gap between the vehicles leads to less air drag, which reduces fuel consumption. In this thesis, two methods for leader-follower platooning control without inter-vehicle communication were simulated and implemented on two Turtlebot3 Waffle-Pi robots. Both controllers only relied on visual information and were prescribed performance controllers. The first controller used the relative distance and bearing angle between the vehicles. The second controller used the pixel coordinates of a feature point in the follower's camera. Experiments were done where the robots drove in different patterns to evaluate the controllers. The original controllers only used visual information. Modifications were made so the controllers could use a camera, LiDAR, or both. A moving average filter was added to reduce noise. The filter caused issues in some cases because it introduced a delay, which affected the system in situations where the velocity was dynamic. Using a LiDAR made the methods more robust toward problems that could occur from using the camera but worse at measuring angles. The LiDAR measurements were noisy, and the LiDAR in the physical system had a bias, but using it was more stable than using the camera in cases where the visual information was lost or affected by external factors. Both controllers worked well in the simulations, with the pixel-based method being better. There were issues with the pixel-based method in the physical experiments due to it being sensitive to external factors that affected the camera, such as the robots driving over uneven flooring. Because of this, the distance-based method was more suited for physical implementation. Adding a LiDAR improved the physical pixel-based method's stability, as it negated the camera issues. If the trajectory of the robots in the experiments was curved, the pixel-based method would shortcut more than the distance-based method.