This project challenged the robot to navigate a maze containing seven objects, making real-time navigation decisions based on what it identified. A Picamera captured live images fed into a Teachable Machine model trained to classify the specific objects placed in the maze. The camera also provided visual feedback during navigation, allowing us to monitor what the robot was recognizing as it moved.
An infrared sensor measured distance to objects, triggering the decision logic when the robot came within threshold range. Upon identifying an object, the robot stopped and executed a 90-degree rotation — clockwise or counterclockwise depending on the classification. The direction mapping for each object was confirmed ten minutes before the presentation, so the code was designed to be quickly reconfigured on the spot.
The system combined sensor fusion — vision for object identity, distance sensing for proximity — with ROS2 action commands for motion control, creating a modular and adaptable navigation pipeline.
