- Giovanni Bernal Ramirez - MAE
- Manasvi Boppudi - ECE
- Yingxuan Ouyang - ECE
- Kevin Nguyen - MAE
- Overview
- Key Features
- How to Run
- Hardware
- Challenges
- Areas of Improvement
- Acknowledgements
- Course Resources
- Other Course Deliverables
This project was to develop a self-parallel-parking robot utilizing only a single camera and 2 2-D LiDARs. It will autonomously navigate until it detects other vehicles and starts to detect and evaluate potential parking spots.
- Object collision avoidance
- Park into parallel spaces
- Ability to park into much tighter spaces by doing multi-point turns
- Reverse parking into perpendicular spots
- Computer Vision: Finds other cars to recognize where parking spaces are available. Helps guides the LiDARs and with vehicle guidance
- Collision Avoidance: LiDARs prevent the vehicle from hitting an obstacle
- Run
all_components.launch.pyinucsd_robocar_nav2_pkgpackage- Sets up all sensors and controllers
- Run
parallel_parking.launch.pyinparallel_parkingpackage
- ROS2 Foxy
- Traxxas 1/10 Ford Fiesta Chassis
- Jetson Nano - Main processing unit
- FLIPSKY 75100 Pro V2.0 VESC - Motor and steering controller
- OAK-D Lite Camera - Computer vision
- DTOF LiDAR LD19 - Rear LiDAR sensor
- SICK TiM571-2050101 - Front LiDAR sensor
Our project relied heavily on CAD for cable management, sensor mounting & integration, and course planning. Onshape was used.
CAD files can be found in the cad folder
Like most projects, this project required a mix of both technical and project managing knowledge.
- Multitude of technical issues that required significant time to debug
- Wi-Fi connectivity was spotty, interrupting SSH sessions that threw away work
- Compatibility issues with certain versions of Python not working with OpenCV while certain versions of OpenCV did not worth with the Jetson requiring rebuilding and specific prerequisite versions.
- Hardware provided had limited documentation, significantly slowing down diagnosing of faulty components. Servos, motors, VESCs, and Wi-Fi adapters had to be replaced.
- Documentation
- Specific hardware documentation was lacking in specifications or troubleshooting requiring guess-work or timely analysis of PCBs.
- Lack of internal team documentation from past-bugs or day-to-day changes caused many small delays where team members would have to ask others for specific settings/commands
- Underestimated time for debug and testing
- Underutilization of human resources
- Software & Hardware projects were often dependent on each other. Looking back, delaying one side to advance the other could be a worthwhile investment. Especially since Hardware is dependent on hours of certain facilities whereas software can be developed almost anywhere.
- Better Spot Evaluation Logic: Shape fitting algorithms & deeper sensor fusion to measure parking spaces more accurately
- Automatic Dataset Collection: Improving the computer vision model to a point where it can help with annotating images
- Dynamic Parking Spot Detection: Differentiates between parallel, perpendicular, and drive-in spaces
- Improved Motion Controls & Path Planning: Better initial position & multi-point turning to get into very narrow spaces
Thank you to the dedicated staff that made this course possible and helping us throughout the quarter
- Professor - Jack Silberman
- TA - Alexander
- TA - Winston
- TA - Jingli
- Contains primary documentation and past ECE MAE 148 final projects