Traction, Slip, and Machine Learning

[Activist_Investor]

Rivian seems like it is a pretty smart car. Does Rivian collect data related to wheel slip and traction? If they do then they could make use of a combination of:

• relative incline
• traction
• terrain identification via image analysis
• exact GPS

Then train a sophisticated machine learning model and provide just in time traction control per-motor. They could also make use of the ride height system per-wheel to aide traction on slip. I bet that would offer a marked increase in on-road and off-road performance compared to traditional traction control systems. It would also just fallback gracefully to on-board standard traction control.

Putting something like that on an R1 and then comparing the off-road performance to a Jeep head to head would be seriously awesome!

Granted I wouldn't spend too much at Rivian on small traction related gains but I would at least collect the data and preferably collect the data specific to precise GPS for future traction models. Notice human behavior is often a counterintuitive thing as well. A Rivian off-road enthusiast may have more fun if the traction is less than perfect (no machine learning or terrain image analysis) and offers an opportunity for the limited mental capacity of a sapient to be made use of to adjust steering.

Anyone know to what extent they are doing this?

Sponsored

 

[zefram47]

This would be kinda pointless. Conditions change constantly, especially in different weather conditions. Snow, rain, mud, wet sand vs dry sand, wet leaves, rocks...everything in between, etc. Current calibration sucks as soon as you lift 2 wheels off the ground, common in off-roading, or wheels wind up in a split-mu situation where one wheel may have lots of grip and one or more do not. Very common and what you're suggesting would do absolutely nothing to resolve issues. Part of the problem is that Rivian calibrated the system in Moab for off-road stuff and Moab pretty much has grip everywhere on the slickrock and that just isn't what most of the country's trails are like. It's also a reactive system where they can't detect the mu of the surface without first trying to apply torque/power which can lead to some undesirable handling characteristics on side slopes and in split-mu situations.

 

[Activist_Investor]

This would be kinda pointless. Conditions change constantly, especially in different weather conditions. Snow, rain, mud, wet sand vs dry sand, wet leaves, rocks...everything in between, etc. Current calibration sucks as soon as you lift 2 wheels off the ground, common in off-roading, or wheels wind up in a split-mu situation where one wheel may have lots of grip and one or more do not. Very common and what you're suggesting would do absolutely nothing to resolve issues. Part of the problem is that Rivian calibrated the system in Moab for off-road stuff and Moab pretty much has grip everywhere on the slickrock and that just isn't what most of the country's trails are like. It's also a reactive system where they can't detect the mu of the surface without first trying to apply torque/power which can lead to some undesirable handling characteristics on side slopes and in split-mu situations.

Do you know if the current system makes use of the ride height system to place a slipping wheel back on the surface?

 

[SaucyMonk]

Do you know if the current system makes use of the ride height system to place a slipping wheel back on the surface?

No, and it really couldn’t. The suspension in these trucks move way too slow to be useful in that capacity. And even if they could, you’d quickly run into an air supply issue. Try raising, lowering then raising the truck again while the compressor is running and see how long it takes to get back up.

 

[ev_fanatic]

Using ML to control traction is a very interesting idea. It would, however, take time and not be very easy to implement as there are a lot of moving parts:
- Getting data for various scenarios/conditions
- Finding compute (presumably this could be whatever driver+ runs on)
- Running the model quickly enough to be useful
- Developing said model consumes developer time, and they just had a layoff

Would be very exciting to see something like that for sure.

 

[R1Thor]

Do you know if the current system makes use of the ride height system to place a slipping wheel back on the surface?

This...is actually exactly what the Rivian does. You can see it in action here:


(tried to paste YT Link with time stamp; didn't work first time--if you can't find what I'm referring to, navigate to ~11:40)

Rivian uses an adaptation of suspension that you can find on Mclaren sports cars, using hydraulics (instead of sway bars) to maintain suspension articulation in order to maintain all 4 wheels on the ground.

 

[Zoidz]

Here is Rivian's current patent that will answer some of your questions.

Sponsored