Simultaneous Localisation and Mapping using LiDAR for Autonomous Racing - Michael Mattiske, 2018

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:

Effective autonomous racing requires an accurate map of the race track and a precise and low-latency estimate of the vehicle's position within the track. This is known as the Simultaneous Localisation and Mapping (SLAM) problem. An end-to-end solution has been developed to solve this problem. The LiDAR sensor was chosen to provide environment perception, available models were investigated and an appropriate unit was selected. Mounting hardware was designed to mount the LiDAR unit to an existing racing car. With the platform this provided, a cone detection algorithm was designed and implemented to allow the system to perceive the cones that demarcate the track. These cones were used as landmarks in an Extended Kalman Filter (EKF) based SLAM solution that was implemented from scratch with adaptations to improve accuracy and reliability. Once the full system was shown to work effectively and provide the required functionality, it was implemented in optimised C++ code, ready to operate on an autonomous racing vehicle in real-time.

Introduction:

The new driverless competition ran for the first time at Formula Student Germany in 2017. At this event only one driverless car managed to complete all competition events. At the end of 2017 several students were recruited to work on developing the autonomous systems for the driverless vehicle and to undertake final year projects on relevant topics. The goal for the autonomous section is to have a car that can complete all events by 2019 with a plan to compete at FSG in 2020.

The competition is split into static and dynamic events. The static events test the team’s business acumen and engineering knowledge, while the dynamic events test the performance of their race car. This FYP is focused on designing a system to contribute towards the dynamic events in the driverless category. These events are the following:

•  Skidpad

• Acceleration

• Track Drive

• Efficiency

The Skidpad and acceleration events test the car’s lateral and longitudinal acceleration respectively. The Track Drive involves navigating a previously unknown track for the duration of 10 laps and is the event with the highest points allocation. Points are scored in the efficiency category based on energy consumption during the Track Drive. All these events must be completed by the car without a driver.

Generally speaking the key challenges involved in converting an existing electric vehicle into a driverless vehicle for the Formula Student Driverless event can be split into three categories:

• Perception: Generate a map of an unknown track and determine the vehicle’s location within this map and its state.

• Planning and Control: Given a full or partial map of the track determine and execute the optimal racing line to minimise time to complete the event.

• Supporting Hardware: Provide the computing hardware, software infrastructure and low voltage power and communications to enable the other categories to achieve their requirements safely and efficiently.

Monash Motorsport is tackling these challenges by assigning engineers into the following subsections which will be working in parallel to design the autonomous system of the driverless car while the rest of the team will be providing and maintaining the 2018 electric car as the platform for the design.

• Cameras

• LiDAR

• GPS/INS

• Path Planning

• Vehicle Actuation

• Low Voltage Systems

• Computing and ECU

Some significant rules changes have been made to the driverless competition for 2019 [1]. The most important of which is the introduction of the Autocross event. This event is a single lap of a previously unseen track. Importantly, data collected during this event can then be used during the Track Drive. This separates the previously combined challenge of navigating an unknown track and racing a known track at high speeds into two events. This serves to remove some complexity from the competition as manual tweaks can be made to the track map between Autocross and Track Drive to fix errors made by the SLAM algorithm. In the short-term this is a wise decision as it should lead to more teams successfully completing the events and results in more exciting spectating.

Redundancy is an important requirement in developing any autonomous system. This is even more the case for an autonomous vehicle capable of driving at high speeds and therefore of causing significant injury. For this reason, multiple sensors are required to ensure that in the case of a single failure the vehicle does not become unsafe. Many vehicle state sensors are available and in fact already exist on the Monash Motorsport cars. However, a key requirement for the event is to be able to detect the cones that act as landmarks demarcating the track. This significantly limits the applicable sensors. The obvious choice is a camera or set of cameras to provide depth information and our car will include such a system. Another excellent option for this use case, while less ubiquitous is LiDAR. 

LiDAR is a 3D range finding technique that uses pulsed lasers to detect the distance to objects in the environment by timing the return of the laser after reflection off a surface. The most common method of providing 3D imagery from this is having multiple lasers arranged in a vertical array and swept across the scene by rotating the device through 360 degrees. This provides some distinct benefits over other perception sensors such as cameras. Firstly, a much greater horizontal field of view is achievable than most sensors as the whole 360 degree range is visible. In addition, the depth accuracy of LiDAR is excellent and does not scale poorly with distance as per stereo cameras. Finally, due to several factors the total sensor pipeline introduces much less latency for LiDAR than with other sensors such as cameras. These factors include the lack of exposure time and lower data rates leading to much faster processing and object detection. Cameras do however offer their own benefits such as colour vision, improved range in our application and well developed and supported libraries for object detection. At Monash Motorsport for our autonomous car we have chosen to use both of these perception sensors to provide redundancy and to augment each other by their complementing strengths and weaknesses.

Conclusions

This project has developed the LiDAR system for an autonomous race car and hence provided the framework for an integrated perception system. The full LiDAR pipeline and supporting hardware has been developed. The LiDAR system provides accurate and low-latency simultaneous localisation and mapping of the vehicle under racing conditions. The SLAM framework developed here will be built upon with measurements from the other perception sensors to form the complete perception system for the car. The progress made in this project contributes to Monash Motorsport being in an excellent position going forwards to continue the development of its first driverless race car.