Hacker News Reader

Comment by rjdagost | Hacker News Reader

A Twitter user who rooted his Tesla claims that autopilot disengagement reports are very small (< 1 KB) and do not contain any actual sensor data, just things like GPS coordinates, speed, heading, etc.: https://twitter.com/greentheonly/status/1096322810694287361

Not the kind of information that you use to train computer vision algorithms. I found his claims to be an interesting read.

View on HN

Definitely interesting but I wouldn’t trust a rooted system to behave the same as a non-rooted system.

Nor is there any guarantee that the snapshot in time when he did his research is representative of everything they do.

And such logging behavior can vary by car, by location, and by software version.

I tend to think it is in Tesla’s interest to collect the most useful data, efficiently, and make good tradeoffs when doing so. They say they are gathering data on interventions that will help improve their neural nets. I believe them.

You might want to read the whole Twitter thread again.

The long section where he describes “campaigns” is Tesla's mechanism for capturing high quality data based on certain triggers, and including pictures.

However the rest of the thread explains that Teslas are "trigger"ed to capture imagery data (in rapid succession, almost like a video) in a variety of situations like being close to pedestrians/cyclists not just disengagements.

Reply on news.ycombinator.com