Issue 37 Unmanned Systems Technology April/May 2021 Einride next-gen Pod l Battery technology l Dive Technologies AUV-Kit l UGVs insight l Vanguard EFI/ETC vee twins l Icarus Swarms l Transponders l Sonobot 5 l IDEX 2021 report

28 Maps API, the Nobil.no public charger API and one for the CAN comms protocol. Different sensors use different comms protocols, often owing to bandwidth requirements, so the Pod has a variety of them, such as CAN and Gigabit Ethernet. Potential conflicts between sensors are resolved by reference to models of sensor performance, measurement accuracy and precision, Hallgren says. These are used with what he calls solid statistical frameworks to form a complete picture of the Pod’s environment and the ‘actors’ that include other vehicles and pedestrians inhabiting it. In the name of safety and redundancy, the Pod’s navigation system uses multiple sensors and sources, along with positioning software written in-house. The GNSS devices for example can handle all the major satellite systems, and use RTK correction data when available, to enhance GNSS accuracy. Hallgren says the software is designed to handle GNSS outages by using data from relative positioning sources to maintain positional accuracy. Overall, the system has an accuracy of 10 cm with respect to the Pod’s intended path. Integrated HD mapping Even when enhanced with RTK and backed up by inertial measurements and odometry, much more is required to enable an autonomous vehicle find its way around safely. In December 2018, Einride therefore announced it would integrate DeepMap’s HD mapping and localisation software into the Pod. Focused on building maps from the ground up for each application and operating environment, DeepMap lists a number of attributes it regards as critical for autonomous vehicles. These include centimetric precision, constant updating to reflect changes in the real world, and seamless integration into the self-driving system. Such maps must also support fast and robust localisation to pinpoint the vehicle’s position in 3D space as well as providing efficient data storage and comms between vehicles and the cloud. The company’s COO Wei Luo says HD mapping is one of the first things a self-driving vehicle company needs, as it is essential to map the operational environment and localise the vehicle within it. She characterises the HD map as essentially a database or layer used by the self-driving software stack for perception and prediction in a manner that is economical with real-time computing power. Luo explains that the 3D HD map – with the support of high-performance sensors – helps the perception system by allowing it to retrieve the locations of known objects such as traffic lights, junctions and pedestrian crossings instead of having to apply computing power to figuring out what they are as the vehicle encounters them. The perception system can then focus sensors and processing power on a traffic light, for example, and compute whether it is on red or green and whether a left/right filter arrow is illuminated. A very precise map drastically reduces the computation needs of a perception system, she emphasises. As the map also holds all the information a self-driving vehicle needs April/May 2021 | Unmanned Systems Technology The Pod was fitted with different sensor combinations during its testing and development, principally cameras, Lidars and radars This visualisation of data from the Pod’s laps of a race track shows the Lidar scans mapping the track and the path-planning software working out a racing line