Safety Validation and Edge Case Testing for Autonomous Vehicles (Slides)
Here is a slide deck that expands upon the idea that the heavy tail
ceiling is a problem for AV validation. It also explains ways to augment
image sensor inputs to improve robustness.
very nice presentation. Since all fatal crashes are rare and unique events, I think they qualify as "edge" cases. there are also many thousands or millions of events every day that human drivers encounter and avoid a crash because they have foresight or anticipation of hazards. Neural networks have no foresight. Therefore, the goal of zero crashes with automated vehicles is a myth.
All comments are moderated by a human. While it is always nice to see "I like this" comments, only comments that contribute substantively to the discussion will be approved for posting.
very nice presentation. Since all fatal crashes are rare and unique events, I think they qualify as "edge" cases. there are also many thousands or millions of events every day that human drivers encounter and avoid a crash because they have foresight or anticipation of hazards. Neural networks have no foresight. Therefore, the goal of zero crashes with automated vehicles is a myth.
ReplyDelete