Obstacle Detection and Object Classification for Off-road Autonomous Vehicle
Francesco Borrelli, Professor
Mechanical Engineering
Applications for Spring 2025 are closed for this project.
This project builds upon an existing 1/10-scale off-road autonomous vehicle platform to explore and operate in unknown and changing environments, including sandy beaches, gravel, and forests. The control stack must make use of onboard sensors (RGBD camera, IMU) to map the environment and plan feasible and cost-effective routes in real-time. The operator will have access to the camera feed on a display, and your goal is to design an obstacle detection and object classification software to help the user understand the environment.
Role: Data collection and annotation: Students will use a combination of open-source datasets and training data provided by graduate student mentors and annotate the data.
Model training: Students will design an image segmentation and classification model based on state-of-the-art models
Deployment and tuning: Students will deploy the model on the platform and tune the model for better performance.
Potential extension to a user interface that reports the objects around the vehicle.
Qualifications: Students must have experience coding in Python, including PyTorch and computer vision libraries. Background in computer vision is preferred. Prior experience with ROS is beneficial.
Day-to-day supervisor for this project: Shengfan Cao, Graduate Student
Hours: 9-11 hrs
Mathematical and Physical Sciences Engineering, Design & Technologies