Intention prediction is one of the challenging (and technical interesting) elements of self driving systems. This video at https://youtu.be/YqwDVaIbjbo shows an example of traffic behaviour in a different culture. In Amsterdam cyclist and pedestrians are sometimes given priority as exception; maybe because the driver sees that the cyclist is having a child on the bike, because the weather is bad, and sometimes a bike takes priority knowing that he/she does not have right-away. These cases are difficult for a self-driving vehicle as "intention prediction" is sometimes based on non-verbal "cues". It can be a small sign, or even a facial expression saying "thank you letting me go", or "I will go anyhow, sorry..".
These cues are sometimes challenging to recognise for self-driving systems. First, the perception system needs the resolution to cover gestures, signs and/or facial expressions. Second, the cue needs to be recognised and/or trained. The video also shows there are differences in cultures that further complicate the learning. It is not necessarily true that following the rules leads to efficient (and acceptable) behaviour in different cultures. When the self-driving vehicle is too hesitant and over-careful, the other traffic might take advantage and the traffic flow could be impacted.
Comments