Perception Integration for an Autonomous Vehicle

Monash Motorsport Final Year Thesis Collection

The Final Year Thesis, is a technical engineering assignment undertaken by students of Monash University. Monash Motorsport team members often choose to conduct this assignment in conjunction with the team. 

These theses have been the cornerstone for much of the team’s success. The purpose of the team releasing the Monash Motorsport Final Year Thesis Collection is to share knowledge and foster progress in the Formula Student and Formula-SAE community.

We ask that you please do not contact the authors or supervisors directly, instead for any related questions please email info@monashmotorsport.com

Summary: 

Georgia Ovenden’s 2019 final year project sets out the function of perception systems in our first driverless FSAE vehicle, M19-D. The driverless class involves an autonomous car fitted with sensors, computing units and actuators. As part of our Autonomous Systems, this project walks us through the integration of perception and SLAM (Simultaneous Localisation and Mapping) aspects of the autonomous vehicle. This project also implemented the integration of GPS measurements into the EKF SLAM algorithm to increase certainty of the car’s position. The accuracy of the vehicle’s position contributes to the accuracy of the map created by SLAM based on detected cones which mark out a track. Probabilistic data association was also implemented using the Mahalanobis distance, which increased the accuracy of the track map produced by the SLAM algorithm. Integrating stereoscopic cameras into SLAM proved to be difficult, as the neural network used for detecting camera cones required more training data. This project also details the custom implementation of an auxiliary program to aid in image labelling to further train the neural network in the future.

Introduction:

Monash Motorsport has been active since the year 2000. Traditionally the purpose of the team was to design, build and race combustion cars in Australia at the Formula Society of Automotive Engineers (FSAE) competition. In 2017, MMS entered a second car into competition, an electric vehicle alongside the combustion car. In 2017, the competition expanded to include a new competition class called Formula Student Driverless (FSD) where Formula Student race cars navigate through known and unknown tracks. These tracks are simplistic in that they are primarily on flat ground, with coloured cones to form a path in an enclosed environment.

MMS started a new section dedicated for research and development for the driverless car in 2018. In 2019, our electric vehicle, M18-E, was adapted with sensors, computing units and actuators to become the first ever Driverless car that MMS has produced, named M19-D. The research areas included low-voltage systems, stereoscopic cameras, LiDAR, GPS/INS, Computing, Path Planning and Vehicle Actuation.

In the first half of this project, Georgia Ovenden was involved in LiDAR selection, LiDAR mount design and implementing Simultaneous Localisation and Mapping (SLAM). In the second half of this project, she was part of the Perception section which includes SLAM, LiDAR, GPS/INS and stereoscopic cameras. The motivation for MMS to develop this car is to compete in Germany and promote interest in an Australasian driverless competition in the future. The motivation for this Final Year Project is to improve the perception and localization abilities of our race car.

Conclusion: 

As part of the Autonomous section, this Final Year Project contributed to the perception systems for our driverless FSAE vehicle, M19-D. The GPS has been successfully integrated into LiDAR-based SLAM. The drift in the SLAM system due to the summation of error over each measurement and distance travelled has been corrected by the integration of GPS. The SLAM system also has more accurately determined cone positions by using the Mahalanobis distance as the data association method without sacrificing noticeable computational power. Team members can use camera annotation software to quickly annotate images containing cones from data gathered at testing. The tool has been integrated with an already existing and popular tool, LabelImg. The template matching tool is therefore easy to use and improves efficiency. The customised camera annotation software has limitations in that it can have false positives which need to be deleted and it requires manual editing if the algorithm could not detect all cones. False positives may occur due to the distance of a cone or different shadowing present in the testing area, in which case the user can easily delete them. As Monash Motorsport's driverless package continues to be developed, this FYP provides a foundation for current and prospective Perception Engineers working on M21 and future MMS vehicles.