Motivation
In a university seminar, I had my first opportunity to combine robotics with AI and computer vision in a practical scenario: making a small robot play soccer autonomously. I was drawn to the intersection of real-time perception and robot behavior planning.
Objectives & Goals
The main task was enabling the robot to accurately detect a soccer ball, navigate to it, and score goals autonomously, employing robust visual detection and strategic planning.
Solution & Implementation
I implemented a hybrid visual detection approach using classical computer vision and deep learning methods. To detect the ball, I applied color masks and circular Hough transforms, extracting the position and estimating distance through geometric calculations. Simultaneously, YOLO detection was explored for identifying goals, opponents, and additional visual markers, though classical CV methods proved most reliable for our controlled conditions.
I also programmed the robot's decision logic, including strategies for approaching the ball, maneuvering around obstacles, and accurately shooting towards the goal.
Results & Achievements
The robot effectively navigated, located, and interacted with the ball, reliably executing its autonomous soccer-playing capabilities. Although COVID restrictions prevented competition against other robots, our testing demonstrated impressive solo performance.
Learnings & Reflections
I gained valuable experience combining classical and modern CV methods in real-world robotics, reinforcing my fascination with intelligent robotics and real-time systems.