Production Q4 chip not available until 2018. I speculate that linking 5 Q3 chips to get to semi-autonomous driving is needed until the Q4 chip is ready.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Mobileye and Tesla
- Thread starter Cosmacelf
- Start date
-
- Tags
- Model S User Interface
Production Q4 chip not available until 2018. I speculate that linking 5 Q3 chips to get to semi-autonomous driving is needed until the Q4 chip is ready.
highlights from recent WSJ article QA: Mobileye Founder Talks Self-Driving Cars - WSJ
Mobileye supplies many of them with chips and algorithms that analyze data to construct a picture of the road and surroundings and help steer clear of danger. Clients include Tesla Motors Inc., General Motors Co. and Volkswagen AG. In October, Tesla offered a “public beta” of auto-steering and autonomous lane-changing softwarepowered by Mobileye’s technology.
Mr. Shashua: In fully autonomous cars you want the configuration to be aware 360 (degrees). In urban surroundings things are very complex. You can’t allow yourself to have blind spots not covered by a camera…We have a production agreement with two, tier-1 manufacturers, for a 2017 time frame. This does not mean that the cars produced with this system will be fully autonomous, only that the sensors installed in these cars would be able to support a fully autonomous mode.
Mr. Shashua: Deep learning is a set of algorithms that teach themselves and get better over time. Systems like that are very good at using context, for instance, by looking at the whole image and working out where a lane can be. A car [using such technology] can find the path even when it’s really hard to do, for instance when no lane marks are visible.
WSJ: Mobileye makes extensive use of “tagged” videos of test rides—manually annotated videos showing where cars are during a driving sequence, as well as things like road signs and obstacles. How does the company use this data?
Mr. Shashua: We have access to a lot of data. Tagging data is part of the preparations needed in order to enable machine learning. This is how the algorithms learn. We have tagged videos of rides for 30 million kilometers [about 18.6 million miles]. This data comes from lots of geographies—Europe, the U.S., Asia. It’s also diversified in terms of visibility conditions and road conditions—it’s really unbiased, and that is something that is hard to produce. We have 500 people in Sri Lanka [annotating videos manually] and a few dozen in Israel. It takes time to create data. The fact that we work with so many auto makers helps. Just one company would have found it hard to produce that amount of data.
Mobileye supplies many of them with chips and algorithms that analyze data to construct a picture of the road and surroundings and help steer clear of danger. Clients include Tesla Motors Inc., General Motors Co. and Volkswagen AG. In October, Tesla offered a “public beta” of auto-steering and autonomous lane-changing softwarepowered by Mobileye’s technology.
Mr. Shashua: In fully autonomous cars you want the configuration to be aware 360 (degrees). In urban surroundings things are very complex. You can’t allow yourself to have blind spots not covered by a camera…We have a production agreement with two, tier-1 manufacturers, for a 2017 time frame. This does not mean that the cars produced with this system will be fully autonomous, only that the sensors installed in these cars would be able to support a fully autonomous mode.
Mr. Shashua: Deep learning is a set of algorithms that teach themselves and get better over time. Systems like that are very good at using context, for instance, by looking at the whole image and working out where a lane can be. A car [using such technology] can find the path even when it’s really hard to do, for instance when no lane marks are visible.
WSJ: Mobileye makes extensive use of “tagged” videos of test rides—manually annotated videos showing where cars are during a driving sequence, as well as things like road signs and obstacles. How does the company use this data?
Mr. Shashua: We have access to a lot of data. Tagging data is part of the preparations needed in order to enable machine learning. This is how the algorithms learn. We have tagged videos of rides for 30 million kilometers [about 18.6 million miles]. This data comes from lots of geographies—Europe, the U.S., Asia. It’s also diversified in terms of visibility conditions and road conditions—it’s really unbiased, and that is something that is hard to produce. We have 500 people in Sri Lanka [annotating videos manually] and a few dozen in Israel. It takes time to create data. The fact that we work with so many auto makers helps. Just one company would have found it hard to produce that amount of data.
Similar threads
