Is a Tesla bought today designed to be software upgradeable to full autonomy?
https://www.tesla.com/blog/all-tesla-cars-being-produced-now...
Highlights:
* The manufacturer has in place and has provided the department with evidence of the manufacturer’s ability to respond to a judgment or judgments for damages for personal injury, death, or property damage arising from collisions or accidents caused by the autonomous vehicles produced by the manufacturer in the form of an instrument of insurance, or a surety bond... That's the key condition. If it crashes, the manufacturer is financially responsible and has to carry insurance or post a bond, just like human drivers.
* The manufacturer shall identify any commonly-occurring or restricted conditions, including but not limited to: snow, fog, black ice, wet road surfaces, construction zones, and geo-fencing by location or road type, under which the vehicles are incapable of operating in the autonomous mode and certify that the vehicles are incapable of operating in autonomous mode under those conditions. Reasonable enough.
* The manufacturer shall certify in the application that the subject autonomous vehicles are equipped with an autonomous technology data recorder that captures and stores autonomous technology sensor data for all vehicle functions that are controlled by the autonomous technology at least 30 seconds before and at least 5 seconds after or until the vehicle comes to a complete stop after a collision That's like the air bag data recorder cars have now, but more comprehensive.
* Certification that the vehicle complies with the “Vehicle Performance Guidance for Automated Vehicles” in the National Highway Traffic Safety Administration’s Federal Automated Vehicles Policy This is the national policy.[2] It's not very restrictive; it's mostly self-certification. This is where there's a big distinction between NHTSA levels 1 (basic auto-brake) and 2 (auto-brake plus lane keeping), and levels 3 and above. Basic concept: at levels 1 and 2, the driver must pay attention and systems must not let the driver stop paying attention. This is the "hands on the wheel" requirement. At level 3 and above, the driver doesn't need to pay attention in autonomous mode, and if anything goes wrong, the manufacturer is responsible.
* The national requirements require a "safety assessment letter" from the manufacturer to the NHTSA answering questions in a few key areas:
Data Recording and Sharing
Privacy
System Safety
Vehicle Cybersecurity
Human Machine Interface
Crashworthiness
Consumer Education and Training
Registration and Certification
Post-Crash Behavior
Federal, State and Local Laws
Ethical Considerations
Operational Design Domain
Object and Event Detection and Response
Fall Back (Minimal Risk Condition)
Validation Methods
* DMV reserves the right to revoke the self-driving license of a system that drives badly.
Passing certification gets a sticker for each car, like the stickers for low-emissions vehicles.
All this is far less restrictive than what's required for light aircraft.
Tesla needs to qualify for a license. Time to put up or shut up.
[1] https://www.dmv.ca.gov/portal/wcm/connect/211897ae-c58a-4f28... [2] https://one.nhtsa.gov/nhtsa/av/av-policy.html
It's interesting to see that the burden of insurance falls upon the manufacturer. I think that's fair during these pilots.
I still think there are a ton of legal questions that need to be addressed though. Should self driving car algorithms be standardized across the industry for safety? Because of the nature of saving lives, should self driving software be open for inspection and review in ways other software isn't? If people modify the software in the car, who is responsible? Can car "owners" legally modify code when safety and lives are of concern?
I had a post that addressed a few of these:
http://penguindreams.org/blog/self-driving-cars-will-not-sol...
>Should self driving car algorithms be standardized across the industry for safety?
Just like every other safety relevant feature of my car; Baseline requirements and tests standardised, implementation open to innovation and competition.
>Because of the nature of saving lives, should self driving software be open for inspection and review in ways other software isn't?
Same rules as the software of my ABS system, or of my airbag: it's up to the manufacturer, in practise nobody cares.
>If people modify the software in the car, who is responsible?
If I modify my brakes or my steering system I'm responsible for any damages caused by this modification. Why would software be different.
>Can car "owners" legally modify code when safety and lives are of concern?
Can I legally modify my breaking systems? Yes, but it has to fulfill the same road safety regulations, and frostin depending on what I do getting all required certifications might be impractical.
That's what makes it all work without detailed regulation. Otherwise, there would have to be detailed technical regulations and pre-deployment testing. This way, the manufacturer is responsible. Period.
Volvo's CEO originally proposed that.[1] Volvo Car Group President and CEO Håkan Samuelsson said in 2015 that "the company will accept full liability whenever one of its cars is in autonomous mode." That's now US policy. Tesla and Uber probably hate that; both have tried to evade liability for bad autonomous driving. GM, Ford, and Chrysler seem to be OK with it; they've all survived big vehicle defect lawsuits and recalls. It's going to lead to much safer cars.
[1] http://fortune.com/2015/10/07/volvo-liability-self-driving-c...
[0] https://www.tesla.com/blog/all-tesla-cars-being-produced-now...
Tesla has a level 4 autonomy demonstration video on their site. Maybe computer vision is superior to LIDAR; not a lot of human drivers with laser arrays on them driving around.
Yes, the software is the hard part. That's why it makes sense to use hardware that makes the job of the software easier. LIDAR tells you how far away objects are; if you're using cameras you have to do that with software instead. AFAIK no one yet knows how to do computer vision with cameras in the sorts of situations a self-driving car needs (though I'm not a vision researcher). That means that Tesla is betting on solving a research problem to get the cars they've already sold to start driving themselves. I always thought that was an insanely audacious business plan.
IMO the most successful implementations are ones that can predict accidents that humans couldn't even see taking place (obscured by other cars, etc).
Such as Tesla'a forward facing radar, highlighted the other day on HN preventing an additional accident participant by detecting a vehicle slowing down ahead that the driver could not visually detect?
Lots of LIDARs, lots of cameras.
"Additional guidance data was provided by a video camera used to observe driving conditions out to eighty meters (beyond the range of the LIDAR) and to ensure room enough for acceleration."
"Stanley was characterized by a machine learning based approach to obstacle detection. Data from the LIDARs was fused with images from the vision system to perform more distant look-ahead. If a path of drivable terrain could not be detected for at least 40 meters in front of the vehicle, speed was decreased and the LIDARs were used to locate a safe passage."
This computer vision processing was performed with 1.6ghz Pentium M processors 11 years ago. Is it so far fetched Nvidia parallel processing hardware can't do better with CCDs than LIDAR?
2) In theory almost everything is "just" a compute problem (just enumerate or search over all possibilities - of course this may not finish for millenia). But if Moore's law says the compute is 20 years away, I think it's reasonable to call it a research problem.
The 100k car/driver thing is a red herring. Steering angle / brake / acceleration info is not the kind of data you need for this task. Big useless data is still useless data, even if it's big.
The right kind of data, if this is to be treated as an ML problem, is a 3D model of the environment (this would be the label), along with camera data (this would be the data). Ironically you need LIDAR to get those labels. It bears noting that the state of the art in visual SLAM is not ML based (it's really difficult to get the "ML" problem I just described to actually work - and it's not for lack of data, Google probably has petabytes of this data from their cars).
Those are labels for a different problem.
Not to say LIDAR isn't essential. I think LIDAR will get cheap enough that it'll just be silly not to have it.
I don't think this term even means anything. The problem can be solved by various mixtures of hardware plus software to varying degrees of adequacy.
On the one hand, we know that it's possible to drive a vehicle with high reliability in a wide range of circumstances with just two relatively low resolution cameras... if those cameras are feeding into a trained human-level intellect. What combination of hardware and software is required to generate that intellect is still unknown.
On the other hand, the problem seems relatively well solved using 3D laser scanners and relatively straightforward software. It works pretty well and just needs to be shaken down in all of the possible circumstances it might encounter. It's also unreasonably expensive.
We seem to be converging on an "easiest" solution which uses a mixture of cameras, radars, ultrasound and maybe LIDAR, using deep learning for image processing. That's just one solution, though.
We probably can, but most of us drive with two cameras with approximately 120 megapixels of feature resolution with highly optimized non uniform pixel density and incredible dynamic range. Micromovements allow the software to boost the effective resolution by about a factor of four, to over 400 megapixels. Image stabilization works in multiple layers and executes stabilization movements within 10ms in the fastest layer.
The human eyes and vision system are certainly one of our more impressive parts, and we have a really hard time replicating their performance.
It's not clear that LIDAR offers any real advantages over a camera suite combined with RADAR.
LIDAR has the same limitations as optical cameras with regards to rain, fog, etc.
As image processing capabilities continue to improve, the case for LIDAR gets weaker.
Actually, no. With appropriate signal processing, LIDAR units can see through rain and fog. The early noise returns from rain and fog can be gated out until the system gets a reflection from something solid. There are expensive LIDAR binoculars that can see through fog. You adjust the minimum range until you're seeing past the fog. So far, automotive LIDAR units don't have that feature, but they could.
That being said, i would imagine said hardware would evolve over time with new releases. It would be nice, and probably safer, if cars could actually communicate with each other, rather than have to rely on their own cameras, to determine another vehicles position. Same with traffic lights, road signs, etc.
that's probably a long way off, although i wonder if Teslas have hardware that would allow them to communicate with one another.
I wonder how useful it would actually be though. I don't think you'd want to involve such a complex a distributed system in a hard-realtime domain like deciding whether you're about to hit the other vehicle, the CAP theorem seems pretty unavoidable there.
This is being tested right now eg. 802.11p/ITS-G5/Vehicle-to-everything (V2X) communication https://en.wikipedia.org/wiki/Vehicle-to-everything
A video of some demonstrations by Ford Motors https://www.youtube.com/watch?v=PF1a-g9suR8
Tesla might already have the hardware needed; this implementation for Linux uses Qualcomm Atheros AR93xx chipset devices operating in the 5.9 GHz band https://rtime.felk.cvut.cz/publications/public/ieee80211p_li...
(https://events.ccc.de/congress/2015/Fahrplan/system/event_at... outlines the current security model - car manufacturers are a small set of organisations that could act as intermediary Certificate Authorities and manage key exchange so messages via these protocols can be "trusted")