Earlier this week Tesla hacker @greentheonly shared some high quality videos from the Tesla Model 3 cabin camera. The footage showed that if Tesla wanted to use the camera as a way to monitor driver attentiveness, it might actually do a pretty good job at it.
It turns out Tesla thinks so too. Green has now shared more video from the cabin camera after gaining access to Tesla’s driver monitoring software currently under development.
According to green, the software uses neural nets to determine the probability of a certain behaviour from the driver. The current list includes things like:
- Phone Use
- Driver Eyes Down
- Sunglasses Eyes Likely Down
- Driver Head Down
- Driver Eyes Closed
- and more
Each behaviour is assigned a probability, and based on the video it appears to already do a pretty good job. In a series of videos a “driver” goes through a number of behaviours and the software is easily able to keep up with each one, adjusting the probabilities instantly and for the most part, accurately.
and for completeness, other detection modes part 1 pic.twitter.com/epXJoAS4BU
— green (@greentheonly) April 7, 2021
There does appear to be some faults with the software that still need to be worked out. The driver was able to use printed pictures of people, including Elon Musk, to fool the system.
You can also use images of others @elonmusk
esp. if you need to make those fake tiktok videos pic.twitter.com/cgRkSJZrnH
— green (@greentheonly) April 8, 2021
As mentioned previously, the software is still under development, so we have no idea how close Tesla is to releasing this to the fleet. Hopefully this will replace the current nag to hold the steering wheel, although according to green it will likely stick around and the two systems will be used in conjunction with each other.
If you have some spare time, below is a 12-minute video showing all the different scenarios that were tested.