The Software Stack That Won the Formula Student Driverless Competition Andres Alvarez1 Nico Denner2 Zhe Feng2 David Fischer1 Yang Gao1

2025-05-06 0 0 4.47MB 6 页 10玖币

侵权投诉

The Software Stack That Won the

Formula Student Driverless Competition

Andres Alvarez1, Nico Denner2, Zhe Feng2, David Fischer1, Yang Gao1,

Lukas Harsch2, Sebastian Herz1, Nick Le Large1, Bach Nguyen1, Carlos Rosero1,

Simon Schaefer1†, Alexander Terletskiy1, Luca Wahl1, Shaoxiang Wang1, Jonona Yakupova1, Haocen Yu2

Abstract— This report describes our approach to design and

evaluate a software stack for a race car capable of achieving

competitive driving performance in the different disciplines of

the Formula Student Driverless. By using a 360° LiDAR and

optionally three cameras, we reliably recognize the plastic cones

that mark the track boundaries at distances of around 35 m,

enabling us to drive at the physical limits of the car. Using a

GraphSLAM algorithm, we are able to map these cones with

a root-mean-square error of less than 15 cm while driving at

speeds of over 70 km/hon a narrow track. The high-precision

map is used in the trajectory planning to detect the lane bound-

aries using Delaunay triangulation and a parametric cubic

spline. We calculate an optimized trajectory using a minimum

curvature approach together with a GGS-diagram that takes

the aerodynamics at different velocities into account. To track

the target path with accelerations of up to 1.6 g, the control

system is split into a PI controller for longitudinal control and

model predictive controller for lateral control. Additionally, a

low-level optimal control allocation is used. The software is

realized in ROS C++ and tested in a custom simulation, as well

as on the actual race track.

I. INTRODUCTION

In the Formula Student competitions, based on extensive

rules and guidelines similar to Formula SAE, student teams

throughout the world design and manufacture an open-wheel,

single-seater race car. Originally consisting of only combus-

tion vehicles, the competition was since extended with an

electric category, and starting in 2017, with an autonomous

category (Formula Student Driverless) as well. Points are

awarded for various aspects, the most substantial of which

are the quality of the engineering design as well as the on-

track performance. One of the most technically challenging

disciplines, Autocross, consists of an unknown, closed-loop

and narrow track of around 200 m to 300 m length outlined

by yellow and blue plastic cones, which must be completed

as quickly as possible without hitting any of the cones.

While on track, any interaction with, or remote control of

the vehicle, is forbidden.

Founded in 2006 by students of the Karlsruhe Institute of

Technology, the team KA-RaceIng developed their 5th au-

tonomous car for the 2021 competition. The KIT21d is shown

in Figure 1. It features a carbon ﬁber-reinforced polymer

1Author and Researcher

2Researcher

1 2 Karlsruhe Institute of Technology and KA-RaceIng e.V.,

ﬁrstname.lastname@ka-raceing.de

†Corresponding author, simon.schaefer@ka-raceing.de

Fig. 1. The KIT21d driving at Formula Student Germany 2021. Photo

credit: FSG Partenfelder.

(CFRP) chassis that is equipped with four electric motors

with a maximum power of 80 kW in total, a 470 V battery

with a capacity of 5.2 kWh, and weighs 214 kg.

II. DESIGN GOALS

After ﬁnishing 2nd overall three years in a row at Formula

Student Germany between 2017 and 2019, our main goal

for 2020/2021 was a 1st place overall at all events. In the

Autonomous System, we focused on two points to achieve

this goal.

Increased robustness in localization and path-planning

The analysis of data collected during the test days and events

showed that our car was regularly on the verge of taking a

wrong turn. The planned trajectory was corrected only in

the last second, meaning we drove at the absolute limit.

To drive any faster without making trade-offs in safety, we

needed a correct trajectory much further ahead. To achieve

this, improvements were needed in the ﬁrst three modules of

the autonomous pipeline:

1) Perception: In 2019, cones were ﬁrst detected at a

distance of approximately 30 m, with the median lying

at around 20 m. We set the goal to increase both ﬁgures

by at least 10 m, while maintaining a false-positive rate

near zero.

2) SLAM: To complete the 40 m perception range goal,

SLAM needed to be able to handle the increased num-

arXiv:2210.10933v1 [cs.RO] 20 Oct 2022

Fig. 2. System Overview

ber of landmarks by utilizing a parallelized architec-

ture.

3) Planning: The generation of a correct path depends on

interpreting the mapped landmarks correctly. Our goal

this year was to evaluate new algorithms and compare

them to last year’s method in terms of accuracy in

difﬁcult situations and computation time.

Increasing average speed

1) On straights: To increase acceleration, we set the goal

of implementing a traction control system.

2) In corners: To use as much of the track width as

possible, the precision of the pose estimation and path

tracking had to be increased. Additionally, torque vec-

toring and active yaw rate control were required to

ensure stability in highly dynamic situations.

III. SYSTEM OVERVIEW

The autonomous system software runs centrally on a

multi-core x86 processing unit (Autonomous Computing

Unit, ACU), which provides the necessary computational

power to run our autonomous system in real time. If cameras

are used, this x86 CPU is complemented by the Coral Edge

TPU machine learning co-processor used for running an

image classiﬁcation neural network. The sensors shown in

ﬁgure 2 are connected directly to the ACU via USB3, Ether-

net or CAN. All actuation values are sent via CAN directly

to the Electronics Control Unit (ECU), which manages the

electrical system of the car and continuously performs safety

checks on the complete system. Figure 2 provides a high-

level overview of the communication in our autonomous sys-

tem. The system is implemented using the Robot Operating

System (ROS) framework in the Melodic Morenia release.

Most components are implemented in C++, except for some

smaller modules realized in Python. The central processing

pipeline starts with perception. The cones detected by the

perception system are processed by the SLAM algorithm,

which localizes the vehicle and builds a map of its surround-

ings. On this map, the target trajectory is planned and then

realized by the motion control system. This whole process is

constantly monitored by the supervisor node that performs

health and sanity checks of the other nodes to ensure a safe

drive. Additionally, our Simulation is capable of testing all

the path planning and control parts of the pipeline outside

of the car, aiding us in ﬁne-tuning the system and reducing

necessary test time.

IV. PERCEPTION

The perception system is responsible for recognizing the

position and color of the cones that deﬁne the race track. The

pipeline takes advantage of the precise location information

provided by the LiDAR. Additionally, rich semantic informa-

tion provided by the cameras can be included if necessary.

A. LiDAR System

The KIT21d uses one Hesai Pandar40P, a mechanically ro-

tating LiDAR operating at 10 Hz, to acquire precise position

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

TheSoftwareStackThatWontheFormulaStudentDriverlessCompetitionAndresAlvarez1,NicoDenner2,ZheFeng2,DavidFischer1,YangGao1,LukasHarsch2,SebastianHerz1,NickLeLarge1,BachNguyen1,CarlosRosero1,SimonSchaefer1y,AlexanderTerletskiy1,LucaWahl1,ShaoxiangWang1,JononaYakupova1,HaocenYu2AbstractThisreportdescrib...

展开>> 收起<<

The Software Stack That Won the Formula Student Driverless Competition Andres Alvarez1 Nico Denner2 Zhe Feng2 David Fischer1 Yang Gao1.pdf

共6页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

The Software Stack That Won the Formula Student Driverless Competition Andres Alvarez1 Nico Denner2 Zhe Feng2 David Fischer1 Yang Gao1

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: