Use the traffic-light to fingerprint "good" configurations during automated mode

[ghost]

I started playing with it. The automated feature is great for exploration but lacks some form to revisit really great moments in the session. You could use the traffic light to help with that as follows:

• You could express all presets picked by the automated session manager as a multi-dimensional feature vector
• And then log configurations that were on "green", "yellow", or "red" lights
• Over time that log can be used to train the random-mode selection to favor or limit yourself within those buckets
• For disjoint sessions maybe even the timeline information may be useful to select various greens at different points in the session

It would need some study if you could cluster these configurations using classic machine learning methods (e.g. discover linkage with clustering and principal component analysis), or feed it directly into a generative network for presets that have been trained on an extensive baseline of signals (e.g. use your current automated wizard configuration to see this).

I have no JavaScript experience, see this more as a point of discussion. It would be a great feature to have.

[fallenangel42]

This is a brilliant idea, and I agree that the traffic light system should also feed some sort of feedback into the automated driving mode. I'm not sure we'll have enough data to do some proper machine learning of any level of complexity, but as a starting point we should incorporate some basic behaviours when it comes to the traffic lights (most important of all is to make the red light reduce intensity immediately). I would like the traffic light system to have the same meaning for riders, regardless of if they are being driven by a human or the automated mode, though.