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

Hypothetical: what FSD hardware would you put on a car?

This site may earn commission on affiliate links.

diplomat33

Average guy who loves autonomous vehicles
Aug 3, 2017
12,558
18,389
USA
I know this forum often talks about FSD hardware and we routinely debate LIDAR or no LIDAR, whether Tesla's FSD hardware is enough, what we like or don't like about another company's FSD hardware. So I thought it might be interesting to play a little game: If you were in charge of designing a FSD vehicle, what sensors/hardware would you pick? And to avoid the obvious answer "I would put everything on the car", you have to take into account costs and power consumption and try to keep your hardware as low as possible.

Here's my attempt:
- 8 cameras with 360 degree coverage (3 front, 2 on either side and 1 rear like Tesla)
- Front, side and rear ultrasonics for 360 degree coverage
- Front, side and rear radar for 360 degree coverage
- FSD computer
- Back up power source and redundant wiring.
- Heating elements to melt ice and snow off radar and cameras
 
  • Like
Reactions: Matias
Main issue I see is the lack of redundancy on the side cameras and inability to have stereo vision which I feel will be necessary for accurate depth perception of objects at distance.

So I'd probably just change to a 16 camera set up. Also ensuring that some of the new cameras are on the front of the car to eliminate all blindspots.

Then you have radars covering all directions for redundancy. The ultrasonics stay for parking.
 
  • Like
Reactions: diplomat33
Main issue I see is the lack of redundancy on the side cameras and inability to have stereo vision which I feel will be necessary for accurate depth perception of objects at distance.

So I'd probably just change to a 16 camera set up. Also ensuring that some of the new cameras are on the front of the car to eliminate all blindspots.

Then you have radars covering all directions for redundancy. The ultrasonics stay for parking.

Yeah, I wondered about stereo cameras too. I know that Tesla says that they have a method for getting distance from a single camera but I agree with you that 2 cameras would be better.

Thanks for the reply.
 
  • Like
Reactions: rowdy
I don't think radar, LIDAR, or ultrasonics will ever be able to navigate safely without working visuals, so I don't think I'd include any of them. Tesla has demonstrated that they've already got quite effective distance algorithms for their cameras.

In addition to Tesla's eight, I'd add the typical other three 360 cameras - mirror mounts and front of car looking down, and maybe BMW style side looking front and rear cameras - these make pulling out of parking safer and can provide redundancy and binocular perspective with the other side looking cameras.
 
  • Like
Reactions: diplomat33
Apparently even HW3 in several Teslas built after April has turned out to be hypothetical instead of real.

I mean, this company never stops to surprise in its ability to surprise us how low they can go and how many announcements and promises thay can miss and silently bait and switch.
 
I don't see how Teslas will ever be able to reliably detect cross-traffic at intersections with the current sensor suite. The front fender cameras are angled backwards and the door pillar cameras are too far back on the vehicle. So at best, there is only monocular vision with no redundancy.

I think, at minimum, there should be side binocular vision and side radar. Surely, whatever Tesla determined was required for the front will be required for the sides as well.
 
Assuming the Tesla Sensor Suite is the starting point.

The first thing I'd do is some experiments with Stereo Vision with the current front facing Cameras that Tesla already has to see if I need to change them. I'd use Stereo Vision to reliably detect generic debris of a certain size. While I was at it I'd make sure the front Cameras had excellent vision at night in the rain.

Then I'd work on an advanced HUD to draw lane lines, and boxes around detected objects. This is so the user can see everything the car is detecting in real time. It would probably be a birds eye view of your car as you drove down the road. Like what Tesla has right now, but a lot bigger with a lot more info.

The next area of focus would be 360 degree down facing cameras to better handle parking lots, and other tight spaces. This could also be brought up on the HUD.

I'd add rear corner radars look further back than the cameras do, and to offer redundancy to the cameras.

I would add a software programmable radio for V2V communication. So I could receive stop light information from stop lights, and I could transmit information to cars in my vicinity about the road conditions.

For the initial couple HW iterations I'd use an inexpensive front only Lidar. This would be used to add redundancy, and to trigger situations where sensor data doesn't agree (for retraining the Neural Network).

I'd add a sensor for driver awareness that was similar to what Super Cruise uses.

For the platform vehicle type I'd pick a platform where I could have a largish SUV/Pickup along with a Cargo/Camper Van. I'd price it at $100K (for the base versions), and $150K for the Camper Van.

For SW development I'd focus heavily on driver assistance. Things like visualizations, alerts for cars doing stupid things near where my car was, and an intense focus on navigational maps. I'd also have things like auto-reporting of pot holes. There really are some areas like where I live where they fix pot holes promptly when reported.

The initial feature would be the "leave me alone" adaptive cruise control. This would set your speed in a way that minimizes disturbances. Basically imagine a really smooth cool driver. Someone who hardly reacts unless he absolutely needs to, and stays away from people for as long as possible.
 
  • Like
Reactions: electronblue
@S4WRXTTCS Not bad at all. I would add front corner radars too, though, for cross traffic — one of which could be a dual forward-looking radar.

One of the things that’s often forgotten in the endless and useless Lidar debate (which I consider more a strawman argument) is that it isn’t the type of sensor that’s the end-game. It is the coverage.

If Tesla had twice the amount of cameras to cover some of their coverage misses, that would be good. Equally 360 degree radars could help. Of course also Lidar, but not necessarily. It is the mix that counts and how successfully you make use of it.

Tesla’s 8 camera, 1 radar, ultrasonics suite is an absolutely minimum coverage effort with existing blind spots and little to no redundancy in many directions. That is it’s biggest downside, not lack of any one particular sensor or sensor type.
 
Last edited:
The first thing I'd do is build a good road and traffic simulator to try out the software. Then get the software running and based on what was needed to do that decide what kind of sensors were required. Then there'd be the issue of the necessary processing and memory capabilities, although I suspect this would all be a bit iterative.

I doubt very much that hardware, sensors or processors are even 1% of the work to get this running.
 
If Tesla had twice the amount of cameras to cover some of their coverage misses, that would be good. Equally 360 degree radars could help. Of course also Lidar, but not necessarily. It is the mix that counts and how successfully you make use of it.

Twice the number of cameras seems excessive to me. You are talking about adding 8 more cameras. I don't think you need that many.
 
Twice the number of cameras seems excessive to me. You are talking about adding 8 more cameras. I don't think you need that many.

Don’t take my point so literally, think about what I’m saying: coverage and redundancy. Adding parking cameras to the nose and redundancy towards the back alone would require several sensors of some type.

Waymo has around 18 cameras in 9 camera modules, 5 Lidars and 4 radars. MobilEye has 12 cameras onto which they plan to add 360 degree radar and Lidar. Both of these mean 20 or more long range sensors.

Tesla has 9.
 
Don’t take my point so literally, think about what I’m saying: coverage and redundancy. Adding parking cameras to the nose and redundancy towards the back alone would require several sensors of some type.

Waymo has around 18 cameras in 9 camera modules, 5 Lidars and 4 radars. MobilEye has 12 cameras onto which they plan to add 360 degree radar and Lidar. Both of these mean 20 or more long range sensors.

Tesla has 9.

I am a literal guy. it's one of my weaknesses.
 
I don’t think anyone knows what the working or optimal sensor suite is for a self-driving car — and I am sure there are many that will end up working great for this — but it does seem Tesla is very low on redundancy and has some coverage blind spots as well. There could be many ways to fix that with many types of sensors and sensor arrangements.