RC Car Physics

[Penguin Fart Racing]

Hello All,

As a retired sciences teacher, I am interested in somewhat odd stuff; hence some Squidgy queries and musings. Does anyone have hands-on with the DYNAMOMETERS from https://www.minipro.com or https://www.mcpappyracing.com . How would you run these tools in a WIND-TUNNEL specifically for ON-ROAD high speed RC cars ? Does data from full-sized cars effectively translate to RC cars ? Is there anyone who has tinkered with servo-activated microprocessor controlled front air dams and rear spoilers that used NACA airfoil shapes ? Any info on software, sensors and interfaces to measure and actually calculate aerodynamic parameters for designing car bodies would be appreciated . Would it be possible to differentially wind suspension shock springs using microprocessor guided memory wire ? Direct synaptic control feeding to a hacked RC transmitter would sure be interesting. Sorry for being WAY-OUT-THERE. Must all be due to CABIN FEVER precipitated by very cold weather and COVID isolation :-[

 

[Jerold]

Hello Penguin and Welcome!

Physics is physics and it still applies no matter the scale. This is empirically this is shown with the RC Planes. They fly like the full scale thing.

Physically, the things that are different, or even weird is the scale and the way materials react.

For instance tires. If you take a 1/10 scale car going 50 MPH it scales to 500 MPH. The "normal" stuff we do on our daily driver doesn't work. This is what your tires look like at full throttle. In this case it's like rock crawling on nobby wrinkle wall drag slicks.

The same holds true of the body. They are typically something like .03", scaling that up it would be a .3" thick piece of plastic working as an airfoil/ air dam at 500 MPH. It's just going to flap around. Sure you can make it thicker but then you are adding weight that scales up as well.

I have not used any dynos nor wind tunnels, but I've wondered about it. I see most people getting feedback on their cars through data logging, either from the ESCs or other equipment.

There is an active spoiler for the Arrma vehicles, they are tied it into the throttle.

Anything is possible given enough time and money. Most receivers have some sort of active vehicle control that compensates for tank slappers. If you are talking about more AI to make decisions, then you need a larger platform mostly batteries. Think washing machine size. Performance wise, even and average driver (full size car) can out drive best autonomous vehicles.

One thing that is a key factor to the RC TX/RX pair is latency. It's very light over head and very direct control. When you start adding layers of SW in there from a Micro you will add hundreds milliseconds of delay for each operation. The smarts would need to be on board.

They best way to do it is via hardware. I personally would investigate using Xilinx Zynq UltraScale+ if you want any smarts in the car. This is what nearly every automotive manufacturer is using, because it's fast and there is no SW in the data path.

 

[Penguin Fart Racing]

Jerold THANK YOU for the info on ARRMA vehicles; will have to look into them. Hope people are not incensed by my mentally meanderings. Your comments are definitely worth musing over. Especially the problem of LATENCY. I wonder if this could be minimized by going to highejr clock and bus speeds for the computers incorporating a 128-bit OS on a 11th GEN fault tolerant multi-tasking, multi-threading CPU array ? Streamlined AI with increased data throughput including GPS and drone telemetry along with some machine code Thaumaturgy could possibly reduce LATENCY to Nano or even Pico second times ! If we could just coral what Einstein called SPOOKY INTERACTIONS ( https://www.mpq.mpg.de/6520992/05-revealing-einstein-s-spooky-action-without-destroying-it ), we might actually be able to build a quantum spin based computer. Doubt that RC Vehicles will ever reach speeds where LORENTZ TRANSFORMATIONS ( http://hyperphysics.phy-astr.gsu.edu/hbase/Relativ/ltrans.html ) would have to be invoked

 

[Jerold]

I doubt people are offended here. It's a pretty diverse bunch and everyone likes RCs. If they don't like or relate to it, they can skip the post. So it's all good. But we have gone off in the weeds a bit, so get your lawn mower out.

Full disclosure, I'm a HW guy. I've dabbled in SW and understand a fair amount of the SW/OS stuff, but it's not in my blood like a real SW guy.

The nature of the CPU and SW is that it is a general purpose tool that does everything just OK. If you have a custom tool (coprocessor), it will do one thing extremely well. Keep that in mind.

There are a lot of reasons doing this will difficult in a general purpose CPU. CPU cycles are fast, but I don't know of any modern processor that can deliver a completed (i.e. pin to pin) instruction in pico seconds.To do this you will need CPU and a dedicate/custom AI coprocessor.

Just a little bit about CPU architecture. A modern processor uses pipelining. Pipelining increases clock speed but can also increase latency. You pipeline to reduce the amount time between cycles and "divides" up the big slow peices into smaller faster pieces. Each tiny little part might run at 3 GHz (330ps period), but then you have to operate them sequentially and those tiny little 330ps parts quick add up to ns of delay end to end. This is nothing compared to getting to/from the IO. Now we are talking 100's of ns or even miliseconds in some cases.

Think of it as a truck load of Amazon package distributed among sports cars on a one lane road. Loading/unloading a truck takes longer but it gets it all there at once. Each car is loaded fast and can driver faster, but is limited by space and traffic. There are instances where both work best.

You might look at something like the Jetson Nano. as starting place. It will be limited by amount of AI you can process. This uses a multi resource architecture where the A57 control the other resources including the GPU (Math engine). However, if you run out of juice you will need to move to a larger device and the large chips are more power hungry than Putin and is still limited by the standard IO.

To me, it's well worth looking at FPGAs (Zynq). The total solution is much smaller, more IO, and more performance per watt than any GPU can get near. The tools now allow you to write C-code, import and work with your training models. There are even Python examples of AI. This is an example of coding your high level system in Python and the FPGA overlay is already done. All you do is feed the FGPA part with the Python. OK it's a bit more than that but you get the idea.

 

[Dan B.]

I’m not offended. I’m embarrassed. You guys are rock stars, and I don’t think I’ve ever been able to say that any technical thread on an RC forum was 92% Greek to me. Jeez guys. Let us know if y’all work this stuff out. But then again, if you do, we’ll probably hear about it on the nightly news. Film at 11?

 

[Jerold]

I’m not offended. I’m embarrassed. You guys are rock stars, and I don’t think I’ve ever been able to say that any technical thread on an RC forum was 92% Greek to me. Jeez guys. Let us know if y’all work this stuff out. But then again, if you do, we’ll probably hear about it on the nightly news. Film at 11?

I won't be on the news at 11 for that. The AI stuff is done everyday, there are thousands of people working on it. Just need the right forums. No matter what solution you use you will need power, lots of power.

 

[Dan B.]

I won't be on the news at 11 for that. The AI stuff is done everyday, there are thousands of people working on it. Just need the right forums. No matter what solution you use you will need power, lots of power.

Ok, so finally you two said something I can relate to…More power!!!

 

[Jerold]

LOL. Build it, break it, fix it. If you are having fun run with it.