comma ai | Driving to Taco Bell with end-to-end ML | Harald Schäfer | COMMA_CON talks | CTO

preview_player
Показать описание

Harald Schäfer social media:
View original video on Youtube:
A drive to Taco Bell

Chapters:
00:00:00 intro Harald Schäfer, CTO: A Drive to Taco Bell
00:00:38 making a superhuman driving agent
00:03:14 three parts end-to-end lateral, longitudinal, navigation
00:04:08 taco bell
00:05:35 end-to-end lateral
00:06:50 end-to-end lateral done
00:07:15 end-to-end longitudinal
00:08:44 easy depth
00:10:40 plug into simulator and it works
00:11:25 fun sidenote, traffic code
00:11:50 is it really that easy?
00:14:50 so what did we spend time on?
00:15:10 the great refactor
00:17:15 rejected ideas
00:17:25 rabbit hole 1: stoplines
00:18:35 rabbit hole 2: depthnets
00:20:15 end-to-end longitudinal done
00:20:40 end-to-end navigation
00:21:10 end-to-end navigation in experimental mode
00:21:29 few more changes for taco bell
00:22:44 taco bell code
00:23:09 taco bell video
00:24:30 just a demo?
00:25:45 so when will this be good?
00:27:48 what's next?
00:28:00 needs to be more end-to-end
00:29:35 what's even more general?
00:30:57 billy and silly the bot
00:31:54 come to robotics hackathon
00:32:42 questions
00:33:03 lane changes blind spot sensors and interior camera
00:34:25 lane change indicator status
00:34:48 team vacation, end-to-end long speed limits
00:35:34 depth prediction in simulator, neural radiance field
00:36:18 model training simulated data, augmented real video data
00:36:53 production ready end-to-end applications for openpilot
00:37:37 technics for end-to-end navigation, zoom level
00:37:59 openpilot disengagement analysis
00:38:30 training data from cars that use openpilot longitudinal

Official communication channels:

How to get a job:
How to collaborate:

Are you interested in openpilot?
Knowledge base:
Check out the code:
Is my car supported?
Frequently Asked Questions:
How to setup openpilot:
Comma Secure Shell:
API Documentation:
CAN analysis tool:
Review and annotate your driving data:
Leaderboard:
Comma Connect App:
Research:

Official George Hotz communication channels:

Follow for notifications:
Unofficial communities and resources:

Thank you for reading and using the SHOW MORE button.
We hope you enjoy watching George's videos as much as we do.
See you at the next video.
Рекомендации по теме
Комментарии
Автор

High quality version uploaded -> youtu.be/wdyMUyhHTOg | from $1250 buy -> comma 3X comma.ai/shop/comma-3x | If you have skill apply on -> comma.ai/jobs | Join the community on github.com/commaai/openpilot | Check if your car is supported -> comma.ai/vehicles | How to setup your car comma.ai/setup | follow the official youtube.com/commaai for more livestreams | Stay up to date by following twitter.com/comma_ai | Support comma.ai at comma.ai/shop.
Chapters:
00:00:00 intro Harald Schäfer, CTO: A Drive to Taco Bell
00:00:38 making a superhuman driving agent
00:03:14 three parts end-to-end lateral, longitudinal, navigation
00:04:08 taco bell
00:05:35 end-to-end lateral
00:06:50 end-to-end lateral done
00:07:15 end-to-end longitudinal
00:08:44 easy depth
00:10:40 plug into simulator and it works
00:11:25 fun sidenote, traffic code
00:11:50 is it really that easy?
00:14:50 so what did we spend time on?
00:15:10 the great refactor
00:17:15 rejected ideas
00:17:25 rabbit hole 1: stoplines
00:18:35 rabbit hole 2: depthnets
00:20:15 end-to-end longitudinal done
00:20:40 end-to-end navigation
00:21:10 end-to-end navigation in experimental mode
00:21:29 few more changes for taco bell
00:22:44 taco bell code
00:23:09 taco bell video
00:24:30 just a demo?
00:25:45 so when will this be good?
00:27:48 what's next?
00:28:00 needs to be more end-to-end
00:29:35 what's even more general?
00:30:57 billy and silly the bot
00:31:54 come to robotics hackathon
00:32:42 questions
00:33:03 lane changes blind spot sensors and interior camera
00:34:25 lane change indicator status
00:34:48 team vacation, end-to-end long speed limits
00:35:34 depth prediction in simulator, neural radiance field
00:36:18 model training simulated data, augmented real video data
00:36:53 production ready end-to-end applications for openpilot
00:37:37 technics for end-to-end navigation, zoom level
00:37:59 openpilot disengagement analysis
00:38:30 training data from cars that use openpilot longitudinal

geohotarchive
Автор

Well done Harald! I've been waiting for this talk ever since the previous one two years ago.

Also...

🎉HAPPY BIRTHDAY 🎉

Obliviatora
Автор

Those 300k line of code on the control plane just mean Tesla is behind comma. True my brother has fsb but prefer the Openpilot ride in my ride for Adas.

JPRC
Автор

Can we make a sky plane and use the curvature of the earth/elevation/angle to know the distance?

leblancexplores
Автор

So basically Tesla is copying Comma AI and going end to end too. Ahhh Elon would never admit it. 🤣

glueautomation-solutioneng
Автор

I'm sorry if i missed it on the website, or some other online reference, but will this work in the United Arab Emirates? Or South Africa (my two home countries.) I'd love to give the comma 3x a go!

dewetdavel
Автор

One of the most capable in the world and one of the thirstyest too

marcosadelino
Автор

It very surprising that predicting the depth do not work.

houssemeddine