- Yes, Lex Fridman here. Thanks for the kinds words on the lectures.
- I can't (or shouldn't) comment on the current Tesla approach, because I both know too much and too little. All I can say is that they have an amazing team. I love seeing good engineering and aggressive innovation (of course, always with safety being the top priority).
- We're using an ensemble approach, so both neural networks for end-to-end steering and for perception-only + controllers + planning. The latter is more reliable, the former is more interesting for research. We're researchers foremost, so want to create the tech of the future as much as create tech that's reliable enough to put on the road today.
- Good question. I am someone who believes that driving is incredibly difficult and we humans take for granted how difficult it is. Officially our trip will be L2, so the driver is always monitoring / in-control. But we will publish all video and data and see how close to 100% we get. The goal is to be open and honest in exploring the role of AI in making a safer and more enjoyable driving experience.
1. The honor is ours for having your presence on this forum.
2. I understand. Tesla does have an amazing team and I'm hoping we owners start seeing a bit more than a crappy EyeQ3 emulation. It seems that changes are afoot. Proof of that is the fact that my windshield wipers have a mind of their own!
3. That is interesting (all I can say, sorry, I'm far from an expert). I wish you the best of luck as you push this field forward. Future saved lives are literally hanging in the balance for driverless vehicles. You, and Tesla, and the rest of the field could sharply reduce one of the leading causes of injury and death in the world (though as you are well aware, some places, like India will be a far more difficult challenge).
4. It seems the simpler, or more innate, the harder it is for AI to emulate. Though I did see that backflipping robot do pretty well recently. And Data was my favorite TNG character (Sorry LaVar!).
All I can say is good luck and get your permits in a row (if needed, who knows what your route is). The unfortunate truth is that Congress has not yet passed unified the laws regarding driverless vehicles and some states, like California did with Uber, despite your official L2, might construe the demo to potentially fall under their self driver laws/regulations. So until this wild west of legal uncertainty is homogenized it is going to be a bumpy ride, even for researchers (which is insane given the massive public benefit at stake). I am trying to get my congresspeople and local officials to move but I'm not naive in that the process will take a long time.
Federally, the SELF DRIVE Act passed the House but has been buried in the Committee on Commerce, Science, and Transportation in the Senate for a long time:
Actions - H.R.3388 - 115th Congress (2017-2018): SELF DRIVE Act
Apparently they've been too busy shutting down the government repeatedly and running up $1T deficits to do their jobs.
Maybe that will get sorted out before Tesla is ready for their cross-country demo (heh).
Good luck and if you crush Tesla convince Elon to license your code so they can get a jumpstart on saving lives. Thanks.