Waymo recently announced plans to bring its robotaxi service to London, marking an important step for the UK’s emerging autonomous transport sector. The company’s vehicles have already accumulated more than 173 million miles of fully autonomous driving. Welcoming this development, the UK government estimated that connected and automated mobility could contribute £42bn (US$56.3bn) to the national economy by 2035. It aims to adapt the regulatory framework to allow more driverless vehicles in UK cities over the coming years. But success will depend on more than new rules or more vehicles; it will depend on reliability. Autonomy must prove not just that it works, but that it consistently makes reliable decisions in tough environments. Autonomous systems must constantly decide what information they can trust. The danger of being confident but wrong A Global Navigation Satellite System (GNSS) receiver produces two outputs. The first is a position estimate: latitude, longitude, altitude and time. The second is an estimate of how reliable that position is. Autonomous vehicles rely on both pieces of information. Navigation software combines GNSS data with inputs from cameras, radar, LiDAR and other sensors. Each source contributes to the vehicle’s understanding of its surroundings. The system then weighs those inputs according to how reliable they appear. The difficulty is that traditional GNSS receivers were largely designed for open environments where satellite signals travel directly from the sky to the receiver. In dense cities, the situation is very different: signals frequently reflect off buildings before reaching the antenna, creating what’s known as ‘multipath errors’. The receiver may still detect several satellites, and the geometry of those satellites may appear favourable. On the surface, the system looks healthy, so the receiver reports a high level of confidence in the calculated position. Yet some of the signals used to compute that position may