A complete design-and-test pipeline of a low cost scaled autonomous vehicle: hardware, sensors, navigation, path planning, and tracking

This article discusses a complete design-and-test pipeline of a scaled autonomous vehicle. To start with the vehicle is modeled using a bicycle model. For path planning, hybrid A* algorithm was implemented. For tracking controller, different algorithms were evaluated, namely, Pure pursuit (PP), Stanley and Linear Quadratic Regulator (LQR). To realize autonomous driving, these algorithms are deployed on a Raspberry Pi-4 board mounted on the scaled prototype with other sensors like Inertial Measurement Unit (IMU), magnetometer and two-dimensional Light Detection and Ranging (LiDAR) sensor. The sensor data was fused together to estimate the vehicle pose and a real-time map of the surroundings was constructed using Simultaneous Localization and Mapping (SLAM) scheme in Robot Operating System (ROS) environment. Next the performance of different controllers was evaluated for the vehicle running on an oval track at 5 m/s. The results demonstrate that vehicle is able to track defined path with reasonable accuracy. Stanley controller achieves best performance with a cross track error eRMS of 30 mm followed by PP (40 mm) and LQR (53 mm). The performance of Stanley controller was evaluated even at higher speeds (10 and 15 m/s) and is able to follow the path fairly well even though tracking performance deteriorates with speed as expected. The results validates capability of this setup to test different control algorithms before deploying them in real world commercial cars.