Bachelor and Master Theses

The goal is to create a robot cart that follows a person, works indoors and outdoors in various lighting, and navigates crowds cost-effectively with obstacle detection. Previous methods like GPS and Ultra-Wide Band (UWB) have limitations in accuracy, indoor use, and require custom hardware. 2D LiDAR is affordable but less effective in crowds.

This thesis proposes using a stereo camera and machine learning, focusing on the OAK-D camera by OpenCV. This advanced camera offers high-resolution imaging, 3D depth perception, and on-device AI processing, ideal for real-time applications. The project aims to integrate OAK-D with the robot for efficient person tracking and navigation, even on low-capacity computers like Raspberry Pi or Jetson Nano. The challenge is to find a model that balances tracking accuracy with the onboard computer's capabilities.

To explore the integration of the OAK-D-Lite-AF camera into a robot cart designed to follow a person. 

Person Detection: Enable the robot cart to identify and follow a person. 

Path Planning: Implement algorithms for the cart to plan its path based on the person's movements. 

Obstacle Avoidance: Equip the cart to navigate around obstacles. 

Real-Time Adaptation: Ensure the system can adapt to changes in real-time. 

Automatic Data Labeling: The camera's object detection capabilities can contribute to the objective of automatic data labeling. 

Self-Monitoring Systems: The real-time adaptation aligns with the goal of creating self-monitoring systems. 

- Knowledge on Artificial Intelligence and Machine Learning.

- Knowledge on programming

- Possible for 1 or 2 students

The thesis is provided by Adopticum, Sweden