2016-05-01 | Leave a comment A video article from the Motley Fool: “Monetizing the Autonomous Car: A Long and Winding Road” refers to a study by the Virginia Tech Transportation Institute called: “Automated Vehicle Crash Rate Comparison Using Naturalistic Data”. One of the speakers on the video, Armun Asgari, at time=4:15, says the Virginia Tech report shows that Google Self-driving cars are already better than human drivers. This is a misinterpretation. The human attendants in the Google cars would have been responsible for averting at least some crashes so that this report is not a man-vs-machine comparison, but one of man-vs-man+machine. Google reported 341 disengagements during the period of the tests, 79 of which the human driver initiated. At least some of those 79 would have resulted in an accident. And the 262 robotics-initiated disengagements obviously required the human attendant to take over. How is that safer? The machine data set is too small to be meaningful. This was explained clearly in a recent RAND report: “How Many Miles of Driving Would It Take to Demonstrate Autonomous Vehicle Reliability?” The professionalism of the Virgina Tech report far exceeds the efficacy of the data comparison. The scorecard for Google driver-takeovers appears to be the best of all current test scorecards. The machine capability was (presumably) improving (significantly?) throughout the collection period while the human capability was likely not improving (especially considering the oft-reported uptick in distracted driving). This actually favours the relative machine score, of course. According to the Fool video, half of Americans “still” consider this technology unsafe. For now, that half seems the more rational. Surely, we will get to a point where AVs are indeed much safer, but not as of 2020, except in highly constrained routes and conditions. On this same video, Rex Moore, repeats an extreme projection: that vehicles currently using 3% of the paved roadway at any one moment could instead achieve 90% concurrent utilization. This is a considerable exaggeration. Incautious reporting of market hype like this confuses the reader and does a disservice to municipal leadership trying to plan. Decision makers who think the AV will be here and pervasive “in the early 2020s” are made to be “deer in the headlights”. How does that help any of us whether we happen to like cars or not?