Uber Metrics for Autonomous Vehicle Performance


(Preet) #1

According to documents that were released to the media, Uber performs an MPI which stands for "miles per intervention" and is one of many metrics that Uber and its competitors use to deal with safety driver interventions when AV's are in full AV mode.

The Uber MPI was used on average once every 13miles, which is significantly more than the Waymo MPI which is closer to 5,600 miles. Uber has so far driven 3 million miles. This might seem ominous, but there are a lot of different factors to determine an MPI, and what Uber might determine is an MPI might not be the same for Waymo.

MPI's can stand for the time a safety driver needs to take over due to software failure, or it can be due to hard braking which is not so much a danger as it is uncomfortable. No matter what the reason for an MPI, according to not the documents that the New York Times received, Uber has not made a lot of progress. However, we can take these documents for granted, since we already know that the NYT is not an Uber fan, and fake news is all the rage these days.

When we compare what Recode reported last year, that Uber safety drivers had to intervene on an MPI average of 0.8 miles, it goes to show that by the end of the year Uber's AV MPI improved exponentially. Again, what we do not have are what are the internal metrics that make up an MPI, and why Waymo can drive for 5,600 miles without intervention? If this is the case, then Waymo starting its Phoenix driving in full AV mode explains how far they have come in comparison to Uber. It also bodes ill for Uber, especially their deal with Toyota, for if Uber is stuck at MPI 13, while Waymo is MPI 5,600, then there would be no reason for Toyota to buy into a much less advanced system.

Uber is under scrutiny ever since it caused the first pedestrian death by a driverless car in Arizona last week. The global AV development community is scrutinizing every detail that comes out from this investigation, since the findings will be crucial in identifying issues that will lead to improved safety procedures.

Uber made a statement to the Time magazine, where they said:

"MPI is not a measure of the overall safety of our testing operations and shouldn't be interpreted as such. Miles per intervention is one of many metrics that we use to track our system's improvement, but without context, it can be one of the least useful. For example, depending on where and how it's tested, the same software could result in significantly different MPI. Additionally, companies may define interventions differently from each other."

I continue to wait for more information regarding the accident and will report it and our observations as soon as I find out more.


(Andrew Martin) #2

MPI is a great metric, it means absolutely nothing since, as the article states, we don’t know what is included in the metric. The fact that Waymo is 5,600 MPI to Uber’s 13, goes to show that something is wring in both companies. Uber has too many MPI’s and Waymo has too few. If both companies used the same MPI then it would mean that Uber is not worth the effort, but if the metrics are different, then we need to know what Uber has included that Waymo hasn’t and is Waymo really that good?