Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Mobileye will launch a self-driving taxi service in Israel in 2019

This site may earn commission on affiliate links.
First of all, You can't run Nvidia on regular cars only on Electric cars. Nvidia Pegasus demands 500 Watts. So its already a non-starter.

large.png


Which confirms my comment. "You can't run Nvidia on regular cars... without any changes to the internals."

You do realize that what you actually said is up there in black & white, right? You can't misquote yourself and say you were right all along.

Regardless, if you actually did click through to the results (did you?) you would see that common alternators range from 80A to 140A @ 13.8V. 500W is 45% to 26% of that capacity. It's not at all difficult to imagine an upgraded alternator to deal with this load, and doing so would be a fraction of the cost of the nVidia chip itself + cameras + LIDAR and so on.
 
The amount of patents that Mobileye has in the self driving space is astonishing.
People talk about the hundreds of companies working on autonomous cars and maps, etc.

But how many of these companies are actually using mobileye patents? or actually have patents for what they are doing? We will never know until 5-10 years down the line when the patent battles begin ala samsung vs apple.

Looks like mobileye is locked and loaded from their patent portfolio alone. If anyone doesn't play ball, mobileye can easily end them!

Patents Assigned to Mobileye Vision Technologies Ltd. - Justia Patents Search
 
Looks like mobileye will collect not just metadata of the static environment and its features for localization (lane and road markings, traffic lights, traffic and road signs, poles, barriers, curbs, etc).

But meta data of dynamic actors (cars, potentially even peds) in order to build an accurate human driving models for its simulator.

Its worth noting that This model is not to drive the car but to predict and simulate how other human drivers behave to help test their driving policy.

Amnon has in the past suggested that you can use both offline (simulated) and online real traffic data.

It's interesting that based on linkedin that BMW has engineers ultililizing GAIL (Generative Adversarial Imitation Learning) Networks.

It's not surprising that Mobileye would use bmw to gather human driving behaviour seeing they have a tight partnership and the fact that Amnon has outlined in the past that its what they wanted to do rather than wait till 2021.

"How do you measure the probability of an accident. I can't do it on a test track. I can't not drive on a test track and say everything is okay. Test track do not reflect the complexity of the real world. Am i going to drive around my block 1 million miles and say I drove 1 million miles and everything is okay? This is what people do by the way."

"One way to validate is to build a generative model of how human drive. Similar to GANs that create realistic pictures. You can create realistic trajectories of how humans drives by collecting alot of data. Using the HD Maps, create a computer game where you have agents driving on realistic roads and the trajectory of their driving paths are mimicking human drivers including reckless human drivers. Then you take our vehicle with our robotic driving policy and you drive in the simulator an infinite number of times (millions) and prove that we don't have accidents."

"How else would you validate this? Take a fleet of 1,000 vehicles and measure how much time i hold the steering wheels? Its all very misleading. Because i can drive in simple areas for 1 million miles and show that i don't touch the steering wheel and avoid going into more complicated areas because i don't want to mess my statistics." - Amnon Shashua

Whether or not this additional collection is going on right now in 2019 BMW model cars is unknown.
 
Last edited: