Page - 53 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Fig. 2. Sensor placement on the vehicle: 1 stereo camera, 2 laser scanner, 3 IMU, 4 steering angle encoder, 5 wheel speed encoder, 6 rotor position encoder, 7 GPS 1) Actuators: TUW Racing had to adapt its vehicle with actuators that still allow for human operation of the vehicle. The brake pedal may not be blocked and the steering must be easily steerable by hand despite the gearboxes and motors that are to be added. • Brake System: Due to the already-optimised brake balance, the brake system is mechanically operated via the pedal. Details in the mounting allow the driver to still press the pedal. An alternative way to decelerate is through reverse operation of the motors, although brake response is faster and conserves energy. The brake system must always be able to stop the car within a maximum of 10 m, even in the face of a single failure in the system, including power loss or any mechanical failure. • Steering: The additional steering motor is mounted to the existing steering strut at the top of the monocoque. With an average-size driver in the vehicle, the car weighs about 240 kg and requires 25 Nm to steer while the car is not moving. This is the force TUW Racing designed for, since it would enable testing with a driver. The competition requires students to design emergency systems in a detailed Failure Mode and Effects Analysis (FMEA). For instance, power or mechanical failure to the brakes or steering must be accounted for with fallback systems. When emergency braking is initiated by remote or failure detection in another subsystem, the vehicle must enter a safe state that simultaneously relies on the actuator’s operation. For instance, a vehicle steering 60 degrees to the left while a full brake is initiated should first steer to the center position to optimise friction on the wheels. 2) Sensors: Sensor placement on the vehicle is shown in Fig. 2. • Camera:TUW Racing selected a ZED6 stereo camera for its visual sensors. It features an opening angle of 90deg and a base line of 120mm, connected via USB3, which enables generation of depth images at a range of 20m.Thecamera ismountedon the topmostpointof the roll bar (1). In order to automate the calibration process and to estimate the extrinsic camera matrix, the team used visual markers attached to the vehicle’s chassis on 6ZED Stereo Camera: https://www.stereolabs.com/ Fig. 3. Image of left camera with visual markers attached to vehicle to estimate extrinsic camera matrix specific locations, as shown in Fig. 3 • Laser Scanner:A laser scanner was placed inside the front wing, at the lowest point, as it is a planar scanner and the cones are relatively short. It is tilted downwards, so that it points at the bottom of a cone from the maximum distance. This counteracts damping influence, which could lead to the rising of laser orientation during accelerations, and the lowering thereof during braking. With theuseof (2)aHokuyo20LXplanar laser scanner, the autonomous system is able to detect cones within the field of view using circle detection and heuristics to filter non-cone circular obstacles, e.g. wall detection. • GPS: For accurate absolute position measurement, a dGPS, provided by a Piksi Multi GNSS module (7), is used along with two beacons placed outside the race- track. The beacons allow for more precise positioning than a common GPS system does. • IMU:The relative movement of the vehicle is measured by a motorsport-grade IMU (3), which measures rota- tion in the yaw axis as well as acceleration in the x and y directions. • Odometry: To accurately determine the front wheels’ speed and distance travelled, TUW Racing uses an inductive wheel spin sensor (5) at each front wheel, and for the rear wheels a rotor position encoder is used for the car’s TUW-Racing-developed motors(6). The steering angle is measured by a rotary position encoder (4)connected to thesteeringshaft.Allmeasure- ments acquired from these devices are used within the software framework which is described in the following section. C. Software In order to integrate and process sensor measurements, ROS is used as the base framework and nodes were imple- mented for the following tasks. • SLAM(SimulationsLocalizationandMapping):The vehicle’sposemustbeestimatedand the race trackmust be reconstructed while driving[8]. • Machine vision:The traffic cones, their positions, and colors, must be detected using multiple cameras and a laser range sensor. 53