Page - 52 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Cup3, and the Crazy Car4 race, which has been hosted by FH Joanneum in Austria since 2008, where students from schools and universities are eligible for participation. Other relevant competitions comparable to the FSD are the DARPA Challanges of 2004 to 2007. The car Stanley [9] won this competition in 2005, but since that time many products in the field of autonomous driving have entered the market. Autonomous control is an important new subfield in the automotive sector as commercial autopilots and driver assistance systems become more and more popular. To classify the different levels of system functionality and compare actual driverless cars, the Society of Automotive Engineers has introduced a classification scheme5. Many commercially available software products fall into classes 0-2, thus only observing driver environment and taking on only limited driving tasks. The rare cases of actual automated driving systems in classes 3-5 can be found in GoogleAutonomousCarsorTeslaAutopilots.Googlemakes heavy use of a multi-beam 3D-laser scanner to understand the entire environment. It also solves problems related to the traffic behaviour of human drivers and it is integrated into external maps and weather services. Tesla uses radar and sonar, in some cases mounted behind the vehicle’s outer structure, to detect individual classes of obstacles, e.g. with radar: moving obstacles; or, with ultrasonic sensors: other cars beside the vehicle. [1] [5] Nvidia presented an autonomous control solution where the system was taught to drive exclusively by RGB camera input [2].MachineLearningwasused tomatchsteering input with camera images, so the vehicle steers based on scenes recognised in the camera view. Nevertheless, the driver has to control the throttle. Anexampleofa frameworkforautonomousdriving ispro- vided by Nvidia DriveWorks. It provides interface layers that allow easy incorporation of new sensors, execute machine vision algorithms and even perform trajectory planning. Ad- vantages are its range of out-of-the-box detection algorithms and the heavy use of graphics-accelerating hardware with perception computation times of within a few milliseconds. An apparent disadvantage is that only Nvidia hardware can be used. Another drawback is that the architecture lacks an actuator control component. The approaches and frameworks in Google’s, Tesla’s and Nvidia’s applications solve a variety of problems associated with driverless cars in traffic that are not relevant to the FSD competition. Accidents, recognising people, lane markings, sidewalks or traffic signs do not have to be handled at all in the FSD. There are no intersections where maneuvers have to benegotiatedwithothercars; for that fact, therenoothercars on the track at all. This year, TUW Racing’s main goal is to design a solid base system as an ideal start for improvements and qualitative increments in upcoming years. 3Freescale Cup: https://community.nxp.com/docs/DOC-1011 4Crazy Car: https://fh-joanneum.at/projekt/crazycar/ 5 SAEReport:https://www.sae.org/misc/pdfs/automated driving.pdf Therefore our solution is based on open-source compo- nents. Due to the fact that TUW Racing initiated the project with experts in mobile robotics, the commonly used Robot Operating System ROS[7] is used for modularisation and communication. ROS enables TUW Racing to interface the required com- ponents of the system, from camera to motor control. There- fore, for speedy development, the team selected hardware devices with drivers already available. In the next section, the hardware and the software frame- works used will be described. III. THE RACE CAR In this section, the competition, the race car’s hardware, as well as the software implemented are described in detail. A. Competition The dynamic component of the race has three challenges: an acceleration race, a skid pad and a track drive. The acceleration race is a 75m-long straight track followed by a 100m straight exit lane. The skid pad track consists of two congruent circles, touching externally, with a diameter of 18.25m. The vehicle enters on the tangent of the circles at their contact point, drives the right circle twice followed by the left twice before leaving via the tangent again. The track drive is a closed loop circuit with an unknown layout consisting of up to 80 m straights, up to 50 m diameter constant turns, and hairpin turns with a 9 m diameter, among other miscellaneous features such as chicanes, or multiple turns. The vehicle must recognise and drive exactly ten laps withamaximumdistanceof500meach.The track ismarked by blue and white striped traffic cones on the car’s left edge and yellow and black striped cones on its right edge. The cones are connected via a high-contrast colored line sprayed onto the road; however, the line may extend out to the side. Additionally, the stop zones on the acceleration and skid pads are marked using orange and white striped cones. The distance from one cone to another along either edge of the track is up to 5 m, and the minimum track width is 5 m, except on the skid pad, where the width is 3 m. The traffic cone layout is predefined in the rules. B. Hardware The total power allowed for formula student cars with any kind of powertrain is 80kW, and for electric powertrains a maximum voltage of 600V at any time is allowed. The other major limitation is the wheelbase of at least 1525mm. Aside from those stipulations, students are free to design their cars’ characteristics as they wish. Both rear wheel hub motors of the TUW Racing car have a maximum torque of 30 Nm at a weight of 3.6kg and a gear transmission ratio of 12:1. The total weight of the car is 160.5kg with a top speed of 30.5m/s, accelerating in 3.1 seconds to 27.8m/s. Fig. 2 shows a rough sketch of the vehicle with sensors. 52