Unmanned Systems Technology 010 | nuTonomy driverless taxi | Embedded computing | HFE International marine powertrain | Space vehicles | Performance monitoring | Commercial UAV Show Asia report

30 The company has also developed its own range of Lidar analysis software for map generation, localisation, object detection, object classification and object tracking. While the exact number of Lidar sensors on the taxis has yet to be finalised, the Zoe currently tends to use five of them. “We have one sensor mounted atop the vehicle, near the geometric centre of the car, and we also have sensors at the corners of the vehicle, which are responsible primarily for obstacle detection,” Iagnemma says, noting that these sensors may also be used in localisation. Each Lidar is mounted on the vehicle frame through the body, using a customised mounting bracket and, much like the other sensors, Ethernet connectivity to each one is routed from the sensor to computers stored in the rear. The roof-mounted Lidar is a Velodyne High Definition Real-Time 3D Lidar (HDL-32E) sensor, responsible mainly for localising the vehicle, although like the others it also plays a role in obstacle detection. The company is currently testing two types of Lidar as obstacle detection sensors, corner-mounted and close to the ground: the Velodyne VLP-16, and the Lux from IBEO. Weighing 1 kg and measuring 144 x 85 mm, the HDL-32E consumes 12 W of power and can operate on voltages of 9-24 V. Its enclosure is IP67-rated, protecting it from dust and moisture – the former a past environmental concern in Singapore, the other a constant one. It can also operate in temperatures from -10 to +60 C. The HDL-32E also has a 360 º horizontal FOV and a -30 to +10 º vertical FOV, across which its 32 Lidar channels are aligned. The roof-mounted unit is capable of generating up to 700,000 data points per second in single return mode and 1,400,000 in dual return mode (an option if operators need to double the data rate), using a Class 1 (eye- safe) laser of 903 nm wavelength with a range of up to 100 m. Measurements are synchronised and time-stamped to 1 ms resolution, and accurate to ±2 cm. The Velodyne VLP-16 Puck is smaller than the HDL-32E, measuring 103 x 72 mm, weighing 830 g and consuming 8 W. It is also on a par with the HDL-32E in terms of enclosure rating, operating temperatures, measurement range and resolution, laser class and wavelength, and horizontal FOV. Sixteen channels operate across a ±15 º vertical FOV, generating 300,000 measurements per second (or 600,000 in dual return mode). The IBEO Lux measures across a 110 º (-60 to +50) horizontal FOV at a range of 200 m, with a 905 nm laser and a data update rate of up to 50 Hz. IBEO’s multi-echo technology allows three measurements to be taken for each laser pulse, with a distance resolution of 4 cm, and distance accuracy is 10 cm. Lidar-based localisation Once this combination of information has been collected, in real time, it is then integrated using a sensor fusion process, and inherently weighted according to levels of confidence in the data. The process outputs a dense 360 º data set – the company’s ‘world model’ – which, rather like the 3D mapping systems common to much of the burgeoning autonomous car world, is a perception-based map of all objects that in this case are relevant to the taxi’s navigation through Singapore’s streets. “Development of the software remains an ongoing process, but the effort to date has required many tens of man-years,” Iagnemma says. The vehicle also makes use of more conventional navigation systems of course. A Sparton IMU collects angular velocity and acceleration data, using MEMS sensing technology with three- axis accelerometers, gyroscopes, magnetometers and a barometer for pressure sensing, all encased in a ruggedised, IP67-rated case. A GNSS module from U-blox is also installed, but unlike other driverless cars, that rely heavily on satellite navigation to determine their positions within a map, the nuTonomy taxi places little reliance on GNSS for localisation. As Iagnemma explains, “GPS systems tend to fail in areas with a lot of urban canyons created by tall buildings – leading to what are called multi-path errors – or in areas with a lot of moisture. Singapore has those characteristics, so it’s actually not a good environment for October/November 2016 | Unmanned Systems Technology IBEO Lux as well as Velodyne VLP-16 Lidars are used in nuTonomy’s taxis