UST034

35 2getthere third-generation shuttle | Dossier in its ODD, and what you have to verify and validate is typically scenario-based.” Testing to SOTIF standards is intended to expose any weaknesses in a fully functioning system that can lead to wrong judgements, van der Zwaan says, citing a well-known accident in which a Tesla crashed because its perception system failed to see a white truck against a white background because sunlight was dazzling the camera. It is to avoid this kind of problem, he emphasises, that 2getthere starts with requirements-based engineering, checks the functional safety of its systems and then looks at scenarios. “We can do that because we deploy at Level 4 in a restricted area, and we can characterise that area,” he says. “We can, to a high degree, predict the types of scenarios the vehicle will have to deal with, but the permutations of those are unlimited so you can never validate a full set and all its permutations.” Testing as many scenarios and permutations as possible, as quickly as practical, is where simulation comes in, although he acknowledges that this approach still relies on scenarios that the team can think of and will therefore miss unknown unknowns. “For that you need a constant feedback loop, including from systems in operation,” he says. Future feedback Setting up a system to capture and process all that feedback is the subject of work that 2getthere is doing in cooperation with ZF. Van der Zwaan describes it as a multi-year programme in which they are setting up simulation environments with hardware and software in the loop on test sites and real deployments with customers, all generating data and feeding it into the validation loop. As part of this effort, 2getthere also has access to the neural network AI that ZF has been developing for many years and which continues to be trained on ADAS (advanced driver assistance systems) as they roll out in millions of cars. With five years of investment and development in the third-generation shuttle, van der Zwaan expects to capitalise on the new design over the next decade, focusing on value engineering to reduce the price while enhancing performance. For example, the company plans to use the vehicles at higher speeds while improving their abilities in mixed traffic operations. “It is about better machine intelligence beyond detection and recognition – intent recognition, for example. That is where we will enhance our software over the coming years,” he says. Unmanned Systems Technology | October/November 2020 Group rapid transit vehicle Passenger capacity: eight seated and 14 standing (maximum) or 12 seated and six standing (standard) Length: 6.044 m Width: 2.104 m Height: 2.784 m Wheelbase: 3.7 m Track: 1.58 m Empty weight: 4.5 t Maximum payload: 1.918 t Maximum gross weight: 6.418 t Autonomy: SAE Level 4 Localisation: magnetic, GNSS, visual landmark-based Sensors: Lidar, calibrated radar- camera pairs, ultrasonic Propulsion: central AC motor, front-wheel drive Energy source: 36.8 kWh lithium- ion battery pack Range: 50 km normal operation Recharging: opportunity or 11 minutes from 30% to 80% Maximum cruising speed: 40 kph Some key suppliers Vehicle engineering, production and assembly: Altran, ASAP Battery system: Akasol Charging system: AEP HVAC: DC Airco/Royal Netherlands Aerospace Centre Computer hardware: ZF Lidar: ZF Radar: ZF Cameras: ZF Ultrasonic sensors: ZF Comms with smart traffic lights: ZF, Siemens Wiring: Cemsy Styling: Zagato Specifications 2getthere’s fleet management software gives supervisors an overview of the whole fleet and enables them to monitor and, if necessary, act on the status of individual vehicles