3 ms·
One approach would be to record 100-to-1000 trials of how human drivers navigate each particular edge case in vehicles instrumented for autonomous driving. Sad
by 100ideas 8y ago
One approach would be to record 100-to-1000 trials of how human drivers navigate each particular edge case in vehicles instrumented for autonomous driving.
Sadly, no one seems to have figured out a way to economically motivate human drivers to do this.
- glaberficken 8y agoThe problem are not the expected edge cases, but the unexpected ones. No amount of hard coded reactions are going to do for those.
- eitland 8y agoIsn't this part of what Tesla has been doing though? At least they sell cars that simultaneously has: - human drivers in demanding situations - sensor and compute package planned for autonomous driving From there it is mostly a question of 1. getting permissions to collect the data 2. finding some efficient way to store -> sift for interesting situations -> magic happens here -> lots of regression testing -> self driving car -> profit! I would think?