Autonomous driving startup Wayve says its new robotaxi service which launched last week in the UK capital will be a key test of its machine learning system which attempts to teach cars how to drive rather than programming them to cope with every road they encounter. The London-based AI company began offering supervised autonomous rides through Uber on Thursday, putting its approach to the test on one of the world’s most complicated urban road networks. The service is initially small. Londoners requesting an UberX, Uber Electric or Uber Comfort can be matched with one of Wayve’s electric Ford Mustang Mach-Es at no additional cost, with a trained, Transport for London-licensed driver remaining in the vehicle to supervise the system. However, the technology behind the cars points to a bigger shift in autonomous driving. Wayve’s AV2.0 system uses an approach known as ‘end-to-end machine learning‘. Rather than relying heavily on high-definition maps and hand-coded rules to determine how a vehicle should respond, the system is designed to learn from driving experience, translating information from its sensors into decisions about how to navigate. The company says that this makes it easier for machines to navigate things like roadworks, temporary speed restrictions, parked vehicles, cyclists and pedestrians. That is a different proposition from earlier generations of autonomous driving technology, which relied more heavily on detailed maps, software rules and predefined operating conditions, requiring engineers to anticipate and programme responses to a wide range of scenarios. “Autonomous driving technology will complement the city’s rich transport network,” said Alex Kendall, Wayve’s co-founder and chief executive. Certainly, London, with its constantly changing roads and unpredictable traffic, is likely to provide an exacting testbed for Wayve’s driving model. The approach also reflects a broader shift in how the industry is thinking about autonomous driving. NVIDIA’s Alpamayo platform,