Autonomous Driving: How far is Tesla now?

[BriansTesla]

A few weeks ago, I would have thought that Tesla was behind Google and that both were a long way from a self driving car. After listening to the Video below from Prof. Amnon Shashua of Mobileye, and in light of the ability that Tesla now has to use the fleet to "train" the system as the AP users drive and correct, I think that Tesla will leverage their hi-def maps to rapidly advance towards autonomous driving. I now believe that a self driving Tesla is not years away but months.

Video below has great information about present capabilities and those for the new EyeQ4® that I believe Tesla will be using soon.

 

[cgiGuy]

Pretty amazing what they're already able to do with just a camera and no other sensors at around 25:00.

 

[tom66]

Pretty amazing what they're already able to do with just a camera and no other sensors at around 25:00.

Humans manage a lot with just two "cameras" but we can make judgements and learn like no other computer can yet do.

For the current state of autonomous driving I think it's likely that will need a LIDAR to be able to do things like negotiate a roundabout and handle merging without a signalled intersection. A single/dual camera system would need sophisticated image processing to pick out depth cues to, for example, judge the which lane a car is in when the lane markings are unclear or as on many UK roundabouts and complex junctions completely missing...

 

[Max*]

They can put the hardware necessary for autonomous driving in new cars, but between getting the software bulletproof and legislation, my guess is still 5 years away from Level 3 autonomous.

 

[electrish]

I live in a moderately hilly region, and my AP "loses track" of the lane markings all the time when going over a hill or going into a valley.
So there is still a lot of work to be done, even when excluding interference from soft objects and complex situations, like intersections.

This will be years, not months!

 

[ItsNotAboutTheMoney]

Years. Far enough in the future not to have it influence any current purchase.

The way I see it, this current technology can already make driving much easier and safer, and by making the difference in effort between patient and impatient driving greater, it will encourage patient driving.

Autonomous driving is a different level of detail, with a need to see more and understand more, and with a much smaller margin of error.

 

[NuclearPowered]

I live in a moderately hilly region, and my AP "loses track" of the lane markings all the time when going over a hill or going into a valley.
So there is still a lot of work to be done, even when excluding interference from soft objects and complex situations, like intersections.

This will be years, not months!

The video presentation showed the test car choosing an appropriate lane and travel path with no visible markings. Another clip demonstrated lane detection with many shadows, which has been demonstrated to be a problem in today's ap.

Current AP is out of date when they launch this eyeQ4. He said it would be with an OEM in Q3 15. Can we assume this is Tesla?

 

[electrish]

The video presentation showed the test car choosing an appropriate lane and travel path with no visible markings. Another clip demonstrated lane detection with many shadows, which has been demonstrated to be a problem in today's ap.

Current AP is out of date when they launch this eyeQ4. He said it would be with an OEM in Q3 15. Can we assume this is Tesla?

While I agree with your initial assessment, I didn't hear in the presentation whether eyeQ3 or eyeQ4 was used for this demonstration (somehow I got the impression that it was still eyeQ3) and since he mentioned that even the older eyeQ3 only used 5% of its capacity for computing these images, I would think that a lot can still be done with this system that's supposedly currently installed on all the AP Teslas. I hope that a lot can still be done from a software perspective (without eyeQ4 or the fisheye camera, for example) to get us to a much higher level of autonomous driving with the current hardware.

 

[mblakele]

At CES in January, Amnon Shashua discussed EyeQ3 and EyeQ4, and mentions Tesla Autopilot frequently.

Highlights for me included a slide describing EyeQ4 with "4 production launches 2017/18". That would fit well with the TM3 timeline, but implies that we won't see big improvements in AP until then. However at about 28:00 he mentioned an earlier production launch that would use three EyeQ3 systems in one vehicle to approximate the capabilities of the EyeQ4: tantalizing.