- Automated Driving Systems
- Automated Vehicles
Methodical Robotaxi Development Approaches Should Be Safer in the Long Run
In the fable of the tortoise and the hare, sometimes the sprinter wins the race, but the slow and steady racer often gets to the finish line first. The problem with being a sprinter is that it’s only part of the skill set needed for a triathlon. Make no mistake, developing and deploying safe and robust automated vehicles is a triathlon where the competitors must compete in multiple areas. Multiple components must be completed for automated vehicles to succeed, including the automated driving system (ADS), the dispatch and monitoring platforms, and basics such as fleet management. This is the fundamental premise behind the long-running Guidehouse Insights Leaderboard: Automated Driving Systems report.
Companies that score well in the Leaderboard are tackling all of the various pieces, assembling partnerships to enable a solid go-to-market strategy when the technology is sufficiently mature. Deciding when that level of maturity has been reached is one of the toughest parts of the puzzle. Humans operate vehicles in a wide variety of circumstances, many riskier than others. More than half of traffic fatalities occur on rural roads, but the business focus for ADS is mostly on urban environments with greater complexity but lower speeds.
Prior to 2020, Americans were driving about 3.2 trillion miles annually with about 6.5 million crashes. That’s about one crash every 500,000 miles, or the equivalent of about one crash every 30 years for the average driver. But these average values are not evenly distributed across all roads, and little data exists about near misses. This is where the contrasting approaches taken by Argo AI and Tesla for developing ADSs are interesting.
Argo AI Takes a Thoughtful Approach to ADSs
Argo AI, which is working on an ADS for Ford and Volkswagen, is taking a thoughtful approach. Like most ADS developers, Argo AI has strict training and operational protocols for safety drivers. It has been testing in multiple cities, including Miami and Austin, for several years. Later in 2021, Argo AI plans to start carrying members of the public for the first time. It has partnered with Lyft, part of that relationship includes the ride-hailing provider sharing some of its data.
Argo AI has a trove of data from accelerometers in the smartphones used by Lyft drivers about both crashes and near misses where the driver had to brake suddenly or swerve, along with precise locations and times. This data can provide a robust baseline about human safety that Argo AI engineers can compare their ADS against. Only when they are satisfied that the ADS is safer than human drivers in real-life scenarios will the safety drivers be removed from the robotaxis.
Other Players Are Testing ADSs in the Real World
For several years, Tesla has provided experimental versions of its Autopilot and Full Self-Driving software to customers. Although many Tesla customers are tech-savvy early adopters, they may be more inclined to take a risk to try a new and unproven product. Normally, the consequences of a failed tech product are comparatively trivial. But out in the world where vehicles coexist with other road users, the consequences can be fatal.
Tesla does have some basis for comparison by operating its software in shadow mode in customer vehicles. But with no restrictions on where the system can be used, the experiment is much less controlled, and it’s difficult to judge the system’s efficacy.
There is no universal solution for developing complex safety-critical systems. A more methodical approach with a robust sensing and software stack is much more likely to yield a truly safe and successful robotaxi system in the long run.