The application scenario is warehouse logistics. The JetBot Pro robot must navigate autonomously in a simulated indoor environment, such as a factory or sorting center, by following guide lines traced on the ground (e.g., adhesive tapes of two different colors) and recognize specific objects.
The objective is to demonstrate the robot's ability to perform a transport or inspection task on a marked route while ensuring operational safety by avoiding collisions with unexpected objects or dynamic elements in the environment, such as pallets or other vehicles.
This project addresses two main tasks requiring the integration of different sensors:
The control architecture will be a Hybrid Architecture combining a deliberative layer (AI planning/vision) and a reactive layer (behavioral obstacle avoidance).
Deep Learning (DL) is required by the project to perform the classification task. A CNN (Convolutional Neural Network) allows for automated feature extraction directly from camera pixels to recognize the specific target object.
Obstacle avoidance is a reactive behavior that prioritizes speed and robustness over planning. Raw LiDAR data is used to control linear and angular velocity.
Abstract commands (linear velocity and angular velocity) will be converted into individual wheel speeds. The See-Think-Act cycle will be implemented in Python within the Webots controller.