When a self-driving car prototype operated by Uber fatally struck a pedestrian in Tempe, Ariz., in March, Uber quickly identified the likely cause in software that caused the vehicle to ignore certain objects that its sensors detected. But a further realization dawned on some executives and members of the team: The rush to develop a commercial self-driving vehicle had led Uber to de-emphasize computer simulation tests that attempt to anticipate how autonomous vehicles would react in millions of driving scenarios.

Engineers at the young simulation program were struggling to thoroughly test the company’s autonomous driving software, in part because of a lack of investment in the program, according to two people with direct knowledge of Uber’s autonomous vehicle unit. That stood in contrast to the process at Alphabet’s Waymo and some other major companies developing self-driving cars, where simulation testing was a top priority.