b'118W H Y C I T I E S C O M E F I R S TThe secret to scaling self-driving technology: testing in diverse, dynamic, and complex urban centers.By DR. BRETT BROWNING Executive Vice President of Product Development and Chief Technology OfficerAt Argo AI,we want self-driving technology to safely serve millions of people aroundplaces our cars have already operated. Likewise, interactions with vehicles, pedestri-the world, and cities are where safety is most desperately needed. Dense city centersans, and bicyclists will resemble scenarios weve already encountered. This theory was are by definition complex and often chaotic environments, and tragically, many citiesproved out when Argo AI began testing operations in our sixth location, Austin, in late are experiencing a rise in pedestrian and cyclist fatalities, despite their lower speed2019. Within a few weeks of having test vehicles on the ground, we were up and running limits. Deploying self-driving vehicles in these areas could help reduce the number ofin autonomous mode. Despite this sped-up deployment, we experienced comparable vehicle crashes, while also making the streets safer for pedestrians and cyclists. Whatsperformance to what we observed in other cities.more, self-driving vehicles have the potential to reduce traffic congestion and increaseThat said, just as mastering simple environments does not necessarily prepare an accessibility and convenience in cities. Theyre also a desirable substitute for personalSDS for dense, difficult urban environments, focusing on a single complex urban envi-car ownership. The potential benefits are everywhere. ronment does not fully prepare the system to enter other complex urban environments This is why were taking a unique approach to testing, one that will give us an abilityquickly. Every city is different in unique and challenging ways. (You can read more to scale our technology in a safer and more rapid way so we can expand to cities acrossabout each citys unique quirks on page 136.) the globe. The multicity approach to developing our self-driving system, combined withWhat counts when training an SDS is not just the number of interactions but also our automaker partners expertise in manufacturing and operations, can best deliver onthe richness and diversity of interactions it experiences. A self-driving vehicle that op-the potential that we all see for this technology. erates safely for many miles on the same roads but never encounters a cyclist weaving In order to build a trustworthy self-driving system (SDS) and expand Argo technologythrough traffic, or a pedestrian walking outside of a crosswalkor multiples of each at to many cities, we are testing on public roads in multiple challenging cities now. This expos- the same timemay be challenged when confronted with these difficult interactions. es our system to a diverse set of complex scenarios, actions, and imagery that help us trainBy testing in multiple cities at once, we ensure that the system encounters these kinds the self-driving system. That diversity comes from the uniqueness of how humans and theirof complicated scenarios at a higher frequency.movements appear to the vehicles sensorstheir differences in shape, physical charac- Plus, each new city has its own unique culture, topography, climate, traffic patterns, teristics, and behavior, including the way they drive, walk, ride, and decide when to yield.and driving behaviors. In order to operate, self-driving vehicles need to understand Furthermore, the variety in the urban infrastructure that surrounds our vehicles produceswhat road actors are doing and what they could do in any given location. A more diverse a significant number of complex situations to which our technology must respond.set of experiences better prepares the system for these possibilities, because it has en-The commitment to testing in multiple cities now enables faster deployment later.countered and learned from past circumstances. Simply put, a diversity of complexity If the initial selection of cities is done well, then each additional city will be similar tomakes the system smarter and safer.FullBook_Mar24.indb 118 4/25/21 6:44 PM'