Last week, a Nissan Leaf outfitted as a test vehicle for an autonomous driving startup called Cruise Automation crashed into a parked car in San Francisco.  According to news reports the vehicle was swerving uncomfortably within its lane, and when the test driver decided to re-take control, the vehicle’s heading at that moment rendered the driver unable to avoid hitting a parked car.

Some news headlines have attributed this to “human error,” which is correct to a certain extent.  If the vehicle was left to drive on its own (and assuming all radar and drive-train equipment were operating optimally), it is unlikely that the vehicle would have crashed.

We see this incident in a different light.  We see it as perfect example of common failures in the user experience of autonomous vehicles.  In this incident, the driver’s perception was that the car was not operating optimally, due to its excessive swerving leading to an uncomfortable experience.  This is similar to a common observation we noted in our UX benchmark of the Tesla Autopilot system.  In our test, many first-time users complained that the car didn’t drive how they would prefer, even though they knew it was driving safely.  If the vehicle is not behaving how the driver prefers it to, then they are more likely to take control – even in situations in which it would not be safe to do so.

This also brings to the forefront the topic of HMI which also continually arises in our autonomous vehicle benchmarks. The vehicle needs to effectively communicate autonomous system status to the driver, but as we outlined in our report on best practices for autonomous systems, current implementations do not always do this in an intuitive manner.  If this Leaf had an HMI that clearly indicated system status, would this crash have occurred?