- Automated Driving Systems
- Tesla
Supervised Full Self-Driving Isn't What It Claims
Continuing the theme of autonowashing from my previous blog, here I examine when full self-driving technology is not what it claims to be. In short, full self-driving does not happen if a human is involved in the driving process beyond entering a desired destination. The implication of full self-driving vehicles is that the electronic brains can handle the entire driving task without an organic supervisor somewhere in the loop. No such road vehicle is known to exist anywhere in the world.
Tesla has sold an full self-driving package on every new vehicle since October 2016, when it debuted Autopilot 2.0. The price and the feature set have changed multiple times over nearly 4 years and currently sits at $8,000; however, not a single Tesla has been delivered to a customer that can fully drive itself without supervision. Tesla CEO Elon Musk has repeatedly claimed that full self-driving software would be feature-complete within a few months. “So this is why I'm very confident about its full self-driving functionality being complete by the end of this year” said Musk on the 2Q 2020 Tesla earnings call. This claim has yet to be realized.
Tesla Might Not Lead the Full Self-Driving Pack
While reaching feature-complete status is an important milestone, it could be argued that a number of other automated driving system (ADS) companies have been there for several years. Waymo, the top-ranked company in the Guidehouse Insights Leaderboard: Automated Driving Vehicles report, arguably achieved this status as far back as 2015. At that time, one of Waymo’s automated prototypes drove a blind man from his home to a destination without a human safety driver on board. Today, most of Waymo’s vehicles still have safety drivers on board and those that don’t are remotely supervised.
Several companies developing automated delivery vehicles, including Nuro and Refraction AI, have test vehicles without safety drivers. Each vehicle still has human safety supervisors. Nuro vehicles are followed by another car with a technician to monitor its behavior and remotely shut it down, while Refraction AI uses someone on a bicycle.
While driving in most highway and urban scenarios without intervention is certainly a significant step toward highly automated or Level 4 vehicles, any requirement for human supervision means that at best such a system would be Level 3 or conditionally automated. Musk regularly claims that full self-driving will be capable of Level 5 when the software is feature-complete, but the definitions of both Level 4 and Level 5 specify that no supervision is required. When the vehicle cannot handle a scenario, it must get itself to a safe, minimum risk condition and call back to base for assistance if needed. Even Musk has acknowledged in his definition of feature-complete full self-driving that human supervision would still be needed for an indeterminate period of time. “So feature-complete means it's most likely able to do (the majority of driving) without intervention, without human intervention, but it would still be supervised.”
If a supervisor is needed to watch and intervene, the vehicle cannot be used for a robotaxi or delivery service. While remote or teleoperation is a capability being incorporated by most ADS developers, it is not a solution that scales to watching all vehicles at all times.
Vendors Should Deliver on the Full Self-Driving Promise
Tesla and other companies should absolutely continue development work systems that are truly self-driving. The potential societal and economic benefits are huge. They should also inform the public of real progress. But absolutely no one should be selling any sort of system that requires human supervision as full self-driving or autonomous and regulators need to be more zealous in enforcing this.