Member Info for TheLongestShot

Well at least we will be able to name drop the Mustang as having SEE but we wont see many.

"The Mach-E will start arriving in late 2020, and Ford will only make 50,000 units globally in the first model year."

The Spectacles video is great, because being a small camera, they have omitted the IR filter, so we get to see that unlike the Porsche concept car with a blank space behind the plastic, we can SEE that the lights are on and we are at home in this horse.

https://www.youtube.com/watch?v=DwQbta914no

JC can you remember if that patent for alternating LED illumination is ours?
We always assumed that the Ford deal was for the F150, but I presume that it is for more than one model?

JC,
I have trawled through the reading list (and so far have resisted the pull of Wikipedia to look up the big words) but with the knowledge that TRI do have both Guardian and Chauffeur, it is certainly possible. However, that does fly in the face of what PM is pushing about a reduction in the cost of R&D. He knows that CFP and others would rather he spend the money on free shares. So he is making a big show on cutting costs.

Ok, back to the technical parts the Mobileye system in your link has two parts. 4 cameras for the ground view (for parking assist) generally not used while travelling at speed. and in their implementation up to 8 cameras providing 360 view, of which I presume the front and rear cameras are configured to provide a narrow view at a high resolution for viewing distant objects.

So we could manipulate the images to provide a single 360 degree image which you could project as a simple cylinder around the car - great for humans to look at, but that would mean throwing away all the useful 3-d clues. So instead, object identification happens on the separate images and is cross linked to create a 3d map of matched points in the images.

Grouping the clusters into "convex hulls" or boxes is a task that is common now, getting done efficiently is the problem.

Ground plane analysis can draw everything that is flat and "drivable" converging lines in front of the car are picked up as lane markings - allows land keeping or alerts for crossing without indicating. As that last Conti patent alluded to you can map ground objects nearest the moving objects to calculate range which gives us alerts on getting to close without relying on radar/lidar. That only leaves traffic speed signs and traffic lights and we have got the features. https://www.mobileye.com/us/fleets/technology/

But of course we are adding on the eye gaze vectors to see what the eye is looking at which prevents unnecessary alerts, or more urgent targeted alerts for what we fail to see.

Would a SEEing system be as well optimised as the current Mobileye system - that I can't say, but Conti, Toyota etc have been working on this for a long time too, so the various combined systems could be great.

Now the converse - if you are a mobileye customer, could you add our gaze vectors on top of mobileye's object model? Or is that not available from their black box ADAS systems without having to delve into their API? (Have a listen to Autonocast #160 with Jack Weast of Intel/Mobileye).

If you were Intel, would you purchase a company to add to your Mobileye product range? Who would be a good fit. What would be the price?

Perhaps Antenna are working hard where aviation and insurance execs hang out, after all the auto makers already know our name. Then they can work on investment firms as next year we will be more in their desired range with proven, growing cash flows. We will have our new US based broker, hopefully working with an up to speed PR firm, whose focus may not be us here at LSE, but the big boys.

Don't forget, Sophie has told us the next dates to look forward to:
7 to 10 January 2020 - CES 2020, Las Vegas
15 to 17 January 2020 - Automotive World Tokyo 2020, Japan

Perhaps there is a plan afoot.

While SEE are waiting in a holding pattern for aviation to land and for VW to drive around the block and wave. Volvo are checking that everything is safe. Toyota are doing what they do and are making everything just the way they want it. There isn't much that PM can do to make this news happen. 

You will need a very big aerial to pick up what our PR firm are doing - are they still learning the client or are they working like swans, busy working where we can't see it?

Is fleet fixed? Well we aren't getting to see the fleet numbers, apart from the subtly hidden implication of $40 million of hardware required for next year's purchases. Not to be sniffed at especially with the significant price reduction. But don't forget insurance. but insurance is boring. How do you sell a jam tomorrow idea like insurance? Most tech investors would have no idea of the scale (and cost) of truck insurance, it is not a car with our little darlings in the back. Think of the cost of insuring a fuel tanker, or a truck full of televisions, or even a refrigerated lorry full of 20 tonnes of your weekly shopping. The costs of the goods mean that the insurance is far, far more expensive than you realise. Now, while you think about that in cash terms, what would you do if I offered to save a chunk of that cost and reduce the chances of the crash, meaning that your truck is making money for more days of the year? Interested?
Too good to be true, maybe next year. but then you discover that it wasn't a dream, you see that your peers are getting on board, that will significantly impact running costs and profit. there are more and more trade articles and other companies are trying to sell this to me now, now there is nothing to lose I will sign up, it saves me capital on the outlay and my insurance costs are reducing. Plus, I remember how we felt after Bob's accident I don't want to go through that again.

Turn that around, my insurance business in't sure about this new fangled stuff, it will take time to get up to speed, talk to legal, work through the numbers with the actuary, are there insurance packages that i buy to cover myself ready for this? Who else is considering doing this in our territory, ok lets trial, lets review. Who can install all these trucks. Need to update our contracts, find installers, change Ts and Cs. Print documentation, train staff and talk to marketing. Then we can talk to the 1/12 of our customers who are ready to contract this month.

These are slow processes. PM wants to show that he has covered many territories and significantly increased the volume due to the new deal if he wants a bang for his buck - then he can show that his $40 million is for the egg boxes for his Golden Geese.

Perhaps Antenna are working hard where aviation and insurance execs hang out, after all the auto makers already know our name. Then they can work on investment firms as we will be more in their desired range with proven, go

JC I have some driving to do so will have time for thinking tomorrow. I shouldn't have joked about the dog eating my homework, the post that I was writing just vanished into the either on my phone.

Perhaps it is a good thing that I didn't get to the CMD. I think we would have been blethering to much

It may all have been a misprint, his skill is :
"extract value FROM the shareholder "

JC I have seen my homework assignment, but the dog ate it

Schlem,
It is Alexander Robinson
@alexanderaviate on Twitter

Here is the key point for me.
"Setting up and calibrating the system for each trainee pilot is more laborious than with eye-?tracking glasses"

I don't think that is required with Seeing Machines in aviation (couldn't find a reference) and it certainly isn't required in cars or trucks.

Just because an academic or an article says something is a first, doesn't mean that it is. They may not be informed in the current state of the art!

S7,
Please don't tell me that my first name isn't really The.

Mike

142 comments so far on the Camera Monitor System rule.

This volume is high and each submission appears to be genuine and personal (no copy/paste campaign).

Curious what group is pushing this to their members to comment ??

Other Comment Counts:
EV sound: 50
Occupant Crash: 20
https://t.co/sJRuxEOUwN

See the thread for more details, should help with Guardian sales!

In a university you can get some cameras and write some software and build your own eye tracking or DMS, it is a good enough lab demo, so you get your degree and some funding to start a company to take over the world.

20 years ago this happened a few times, recently with improved cameras, computers and open source software it is happening more and more, so now there are 40 companies doing this in china.

But to move from lab demo to automotive quality means that it must work, not just in the lab some of the time not in the car most of the time, but you have to work almost all of the time today, and in 10 years time you still have to be working with all the variability of people and with degrading components and variable maintenance standards.

Simple DMS looks for yawns, eye closure, head dropping if it can't see your eyes. You can do that in 1 camera with lots of data to train it add more data and it gets better - up to a point.

The next level needs LEDs and an eye model to use glints to track the eyes. It can work, but for improved performance and accuracy you need the whole data path and optics understanding, and PATENTS. If you don't have them you need to work around them.

The next level of accuracy will place the head in 3d, again you need more knowledge and funding to test and build and again PATENTS.

Each level of improved accuracy and scope requires a big jump in funding and IP. That is why AVCC is so important, we are working collaboratively with many Tier 1 companies each with their own IP, but we can all build on and share the IP if we collaborate. That builds a massive moat around the level of accuracy and utility that our DMS and eye tracking can add to member companies. If you are not in the club, you can of course license some of the IP, but it will cost you - but that might get you last year's product (add 2-3 years to release it), but we are already developing the generation or two ahead.

We are the winner in this space, we will take the lion's share, we will take the high end with its premium value , we will take the mass market (where we choose to quote) because we can guarantee performance and accuracy and long term viability. of our solution,. We will be in AVCC members' cars when the time is right.

Thanks JC for all the patent research and helping us see this.

Thanks B,
So they re-ran an old article nothing new there other than the Tesla orange reference.

https://www.wired.com/story/your-cadillac-can-now-drive-itself-more-places/

I can't see the whole article, but looks like they refreshed it this month

Opinion - Afraid, very afraid.

JC,
I felt OK when seeing machines was keeping me safe when driving. But this excerpt from Toyota (TRI) is for building Godzilla sized robots that can see a truck, judge the distance to truck, then "handle" the truck - is that how they power it? Trucks for breakfast?

[0012] Systems, methods, and other embodiments for stereo depth determination are disclosed herein and can be used in many robotic and/or vehicular applications where an accurate distance is needed or desired. As one example, the described arrangements can be implemented on a vehicle and used for detecting a distance to an object in an environment. An object is a thing that forms an element of or constitutes the subject matter of an investigation. Example objects detected by the vehicle include a car, a bus, a truck, a trailer, a pedestrian, an animal, a road sign, etc. The vehicle can then determine a travel path based, at least, on the distance to the object. The vehicle can then control one or more vehicle actuators to implement the determined travel path. As another example, the described arrangements can be implemented on a robot, and can be used for detecting a distance to an object in the environment. The robot can then determine one or more maneuvers based on the object. The maneuvers can include, for instance, handling the object, avoiding the object, etc. The robot can then control one or more robotic actuators to implement the determined maneuvers. While the detailed arrangements discussed below are described in connection with a vehicle, one skilled in the art can apply the invention to robotic apparatus. Moreover, the described arrangements can be used in other advanced control apparatus for controlling actuators.

JC, it just goes to show how sone things can be beautiful when performed, but when described in Japanese / "Patentese" then machine translated to English they lose some of the immediacy and fluidity that they are describing.

So these patents (I am exhausted after reading one) are describing an autonomous car watching to see if a driver of a car, joining at an intersection has seen us! Woah that is a big jump from what I was expecting.

So here we are back to 1 or 2 - find a person, their face and if there are enough pixels, perhaps find their eyes to get best guess on where they are looking. Only, now we have to do this in the big bad world outside of our vehicle!

So hopefully we can get some shortcuts. Find the vehicles, know where to find the driver, "is there a seeing wheel!" that must be the driver!

Schlem,
If it is in Aviation - I spy something beginning with S
"Sky", yes that will be a fun labelling task.
Also they may need a pilot's license rather than a driving license!
;-)

Thanks JC, I knew you would be able to tie it back to individual patents.

Some use cases for 7 and 8:
There is a long straight road with a sharp bend at the end where the bend is marked by chevron signs that are facing dead on so they give a strong reflection to the car's bumper detectors. The car doesn't know that the road is starting to climb uphill and that the corner has a good camber - so I hardly need to brake, as a result if I haven't braked as I approach the bend, the car beeps and flashes warnings. If the car knew that I had seen the sign and was scanning road around the bend, it would perhaps not need to warn, especially if it knew the incline and camber.

In another location, there is a steep hill with few chances to overtake afterwards. So the car will alert me if it sees a slow moving bike and I haven't yet applied the brakes. Here Toyota's Guardian may realise that I need the speed to overtake on a hill and I have not pulled out as I am waiting for the oncoming car to clear.

I was trying to get my head around how much of this problem we need to solve and give it some historical perspective

1. The early days - find body and head, look for slumping. Find eyes, check they are open and by how much, perhaps detect
yawning. Use face position and for rough gaze direction
2. Enough pixels in the camera - can now attempt eye tracking by looking where the pupils are relative to the eye to determine gaze direction.
3. Glints from LED - use the glints and detailed knowledge of eye ball shape to detect where the eyes are actually looking with a high degree of confidence.
4. Map the Gaze to internal or external - Need to have knowledge of the vehicle and window locations. If you can place the gaze source within the 3-d vehicle cabin then see if the gaze aligns with the windows
5. Self mapping of the vehicle space - build your own map of eye gaze locations and classify based on simple rules - expect to see front window, rear view mirror, side mirrors, side windows, instruments , central console, controls on steering wheel, etc
6. Simple match of gaze direction to external changes - validate attention by observing the eye following changes in the external environment - particularly objects moving laterally. Gaze convergence?
7. Object identification - This is where it gets more complicated (and compute intensive) Run an object identification engine on the forward camera, classify the objects and ensure that the eye is looking at them regularly and alert if they do not
8. Object identification with planned route - assuming that we have a map and or a planned route that can be merged with the external view - add more resilience to the object identification and reduce false alerts.

I am no expert in the field, this maybe common knowledge (or still maybe still patent-able?). But you can see that there is a potential path for SEE to follow. I see that 4 would work in mass market, but 5 would be required for aftermarket installations, but could benefit cars too. 6 is perhaps possible today with our existing camera data being used for training and validation.

However, when you get to 7 and 8, you are starting to duplicate the computing effort that I would have expected to be performed by the car. This could be a plus point as we are a separate safety system that may pick up different risks externally and ensure that the driver is made aware, it could even inform the car's system of objects that the driver may perceive that it did not see. There is a strong correlation with the work with TRI where the "Guardian/Chauffeur" system would be the brains, but I did not see us needing to to so much external vision work, but in our Guardian after market product I can certainly envision 5 and 6 happening in the next 2-4 years.

As for bDMS - I am not sure that our cycle time is short enough, but perhaps elements of 5 and 6 could be available sooner - but having this available in the Fovio chip in less that 12 months may not be

OK, I have reversed my position on this. In the past I thought that SEE would focus on the world inside the car. Whereas China and elsewhere have lots of startups with recent graduates using a webcam and machine learning to build DMS, or external vision systems. Many succeeded in building systems that were not well enough trained or engineered to work in all situations (lighting, ethnicities, glasses etc). Whereas SEE started with optical path and years of fundamental research, and has 2 billion miles of data to train on, so has a far more effective system.
Now I believe that they are preparing for Gen 3 and Gen 4, ... of Guardian or the chip where they will fuse the eye gaze and the external world to check that the eyes are sensing and tracking external objects; that the eyes have seen objects that are entering the path of travel, or if the objects ahead of the vehicle are growing rapidly, or change in form and could be an immediate or future risk.
Despite PM saying that R&D would be scaled back, we are still progressing and entering new areas to ensure that we keep 1 or 2 steps ahead of the competition.