The 65th annual Daytona 500 will take place on February 19, 2023 and for many this elite NASCAR event is the pinnacle of the car racing world. For now…
Overview
The article discusses the development of an AI-powered autonomous miniature race car by students at UC San Diego as part of their Triton AI Data Science capstone project. It highlights the use of camera-based navigation as an alternative to Lidar, detailing the technologies and methodologies employed for mapping, localization, and obstacle avoidance.
What You'll Learn
1
How to implement mapping and localization using RTAB-Map SLAM
2
How to use Detectron2 for object detection and avoidance in autonomous vehicles
3
How to fine-tune camera settings for varying light conditions in autonomous navigation
Prerequisites & Requirements
- Understanding of autonomous vehicle navigation concepts
- Familiarity with ROS and computer vision libraries like Detectron2(optional)
Key Questions Answered
What technology was used for mapping and localization in the autonomous race car?
The team used the RTAB-Map SLAM algorithm for mapping and localization, which is compatible with their sensors and allows for real-time mapping without the need for GPS, which can be unreliable at high speeds.
How does the team achieve obstacle avoidance in their autonomous race car?
Obstacle avoidance is achieved using Detectron2, which processes image input to extract information about obstacles and boundaries, allowing the race car to determine its optimal next move while navigating the track.
What hardware components were utilized in the autonomous race car project?
The project utilized a JetRacer with a Jetson Xavier NX developer kit, an Intel Realsense D455 depth-sensing camera, and a Flipsky VESC speed controller for movement commands, enabling effective navigation and control.
What challenges did the team face when using a camera instead of Lidar?
The main challenges included the need for special programming to handle varying light conditions and the requirement for multiple cameras to achieve 360-degree vision, which adds complexity to the system.
Technologies & Tools
Hardware
Jetracer
Used as the platform for the autonomous race car.
Hardware
Jetson Xavier Nx
Serves as the developer kit powering the autonomous vehicle.
Software
Rtab-map
Used for real-time mapping and localization.
Software
Detectron2
Utilized for object detection and image segmentation.
Software
Ros
Framework used for robot software development.
Hardware
Intel Realsense D455
Depth-sensing camera for collision avoidance.
Hardware
Flipsky Vesc
Speed controller to manage movement commands.
Key Actionable Insights
1Implementing camera-based navigation can be a cost-effective alternative to Lidar in autonomous vehicles.This approach allows developers to create smaller and more affordable autonomous systems while still achieving effective mapping and localization.
2Utilizing RTAB-Map SLAM can enhance the reliability of navigation in environments where GPS is unreliable.This is particularly important in high-speed scenarios, such as racing, where accurate positioning is crucial for performance.
3Fine-tuning camera settings for light sensitivity can significantly improve the performance of computer vision systems in dynamic environments.This adjustment is essential for maintaining consistent object detection and navigation capabilities, especially in changing lighting conditions.
Common Pitfalls
1
Relying solely on GPS for navigation can lead to inaccuracies in high-speed environments.
GPS signals can be unreliable, especially in fast-moving scenarios like racing, making it essential to use alternative methods like SLAM for accurate positioning.
Related Concepts
Autonomous Vehicle Navigation
Simultaneous Localization And Mapping (slam)
Computer Vision In Robotics
Ai-powered Object Detection