RobotSports has published two projects on artificial intelligence on GitHub for use by the wider robotics community.
Both projects are written in C++, and a Python API is available as well. The projects are licensed under the permissive MIT license, which makes it easy for others to utilize and/or modify the code.
The first project is a library that implements object detection with YOLOv5 on NVIDIA-based systems.
This project can be found on github: YoloV5-TensorRT project.
Important considerations when developing the library were real-time use on resource-constrained systems (i.e. with limited power usage, and which also fit on the robots) and robustness. In addition, the Github project also contains several examples and detailed documentation. The library, along with YOLOv5, is used by RobotSports for recognition of robots, goalposts, balls and people on the football field:
At the same time, the library can also be used for object detection in other application areas, for instance:
The second project demonstrates the use of YOLOv5 with the Stereolabs ZED - a stereo depth camera.
This project can be found on github: zed-yolov5 project.