Great visualizations of the accident in San Francisco. I'll see you as the expert witness in court. Also great comments on the role of neural nets (should they be ubiquitous or not?) - Cruise I think was quoted as saying their vehicles are engaged in "continuous learning" which usually means NN. Finally - as we move into this era of automation, the key issue to acknowledge in evaluation of accidents: Such inevitable events won't be the same ones we would see in a universe of all-human drivers ... the question is "are we safer?" The metric is severe injuries and fatalities per year. Looking at any particular accident and saying "well that would not have happened with a human driver" is clearly the wrong standard.
Update: Cruise issues recall update to the AV fleet to update the minimal risk response to fail-stop when engaged in a collision. Report linked below. The pedestrian was on the ground after they were propelled in front of the Cruise vehicle after a hit and run incident occurring in the adjacent lane. The Cruise vehicle braked hard and veered to the right when it ran over the pedestrian, registering the collision as a lateral collision versus forward collision. This induced the vehicle to engage in a maneuver to pull over to the side of the road (“fail-op” vs “fail-stop” as outlined in DRIVEn podcast episode 0005). The update will be to make the vehicle fail-stop under such a circumstance. static.nhtsa.gov/odi/rcl/2023/RCLRPT-23E086-7725.PDF
Cruise’s Continuous Learning Machine Predicts the Unpredictable on San Francisco Roads (Cruise): medium.com/cruise/cruise-continuous-learning-machine-30d60f4c691b A Flexible, Verifiable, and Validateable Approach to Autonomous Vehicle Safety (Paul Perrone): www.dependability.org/wg10.4/ivds2021/presentations/perrone-PerroneRobotics-IVDS-21-Flexible%20VandV%20Safety%20Approach.pdf
Great visualizations of the accident in San Francisco. I'll see you as the expert witness in court.
Also great comments on the role of neural nets (should they be ubiquitous or not?) - Cruise I think was quoted as saying their vehicles are engaged in "continuous learning" which usually means NN.
Finally - as we move into this era of automation, the key issue to acknowledge in evaluation of accidents: Such inevitable events won't be the same ones we would see in a universe of all-human drivers ... the question is "are we safer?" The metric is severe injuries and fatalities per year. Looking at any particular accident and saying "well that would not have happened with a human driver" is clearly the wrong standard.
Update: Cruise issues recall update to the AV fleet to update the minimal risk response to fail-stop when engaged in a collision. Report linked below.
The pedestrian was on the ground after they were propelled in front of the Cruise vehicle after a hit and run incident occurring in the adjacent lane. The Cruise vehicle braked hard and veered to the right when it ran over the pedestrian, registering the collision as a lateral collision versus forward collision. This induced the vehicle to engage in a maneuver to pull over to the side of the road (“fail-op” vs “fail-stop” as outlined in DRIVEn podcast episode 0005). The update will be to make the vehicle fail-stop under such a circumstance.
static.nhtsa.gov/odi/rcl/2023/RCLRPT-23E086-7725.PDF
Cruise’s Continuous Learning Machine Predicts the Unpredictable on San Francisco Roads (Cruise): medium.com/cruise/cruise-continuous-learning-machine-30d60f4c691b
A Flexible, Verifiable, and Validateable Approach to Autonomous Vehicle Safety (Paul Perrone): www.dependability.org/wg10.4/ivds2021/presentations/perrone-PerroneRobotics-IVDS-21-Flexible%20VandV%20Safety%20Approach.pdf