Autonomous Ros Car
The goal of this project was to build and program an RC car to autonomously navigate and race a preset course
Rally Car Hardware
Modified Traxxas Rally Model 74076-1
LiDAR
IMU
Arduino
NVIDIA Jetson TX2 single-board computer
Software:
Our rallycars are equipped with NVIDIA Jetson TX2 single-board computers running an upgraded Ubuntu 20.04 OS. This setup features ROS noetic, enhanced with CUDA 8.0, TensorRT 8.2.1.8, and a custom-built OpenCV 4.2 with CUDA support to leverage GPU acceleration for advanced processing tasks.
Speed Control:
The adjust_speed function modulates the car's velocity based on upcoming turns and changes in the angular trajectory. By looking ahead at a set number of waypoints, it assesses the angular changes and distances, slowing down for significant turns to ensure safety and stability, and speeding up when the path is relatively straight. This allows for smooth and adaptive navigation by ensuring the car reduces speed safely before sharp turns and maintains optimal speed on straight segments, enhancing both driving efficiency and stability.
Steering Control:
The drive_to_waypoints function navigates the robot vehicle through a sequence of predefined waypoints by utilizing a PID controller to adjust steering. Running in a controlled loop, it calculates the heading and distance to each waypoint, dynamically adjusting the car's speed and steering to ensure accurate path following. The function continuously publishes the necessary control commands and checks if each waypoint is reached before proceeding to the next one. This allows us to autonomously navigate the vehicle along a predetermined path with precision, ensuring it effectively reaches each waypoint while dynamically adjusting its speed and steering for optimal performance. The continuous loop and real-time adjustments provide a robust method for smooth and accurate navigation.
Localization
AMCL (Adaptive Monte Carlo Localization) is used to refine the car's position estimate by integrating sensor data within a known map, leveraging probabilistic algorithms. Odometry provides continuous estimates of the car's movement based on wheel encoder or IMU data, which is essential for tracking the car's position and orientation over time. Hector mapping is utilized to build and refine the map of the environment without relying on wheel encoders, using high-frequency laser scan data for enhanced accuracy. Together, these components ensure robust and precise localization for autonomous navigation.
Fully Autonomous
Navigation
The Result
The final revision of the code could constantly set a lap time of 36 seconds with the final official lap time of 35.76 seconds placing us in first on the podium. Just 1.2 seconds off the 7 year standing lap record of 34.5 seconds. With more time and tuning I believe we could have broken that record.
