Just heard about this, this morning…
Just heard about this, this morning…
I’m intrigued. I had the idea to program a PID controller for my watercooling system for a long time now. This would be bloody overkill just like I like it …
On the other side a H265/264 decoder comes in handy for streaming too!
I used to be all hyped up about new versions of the raspberry pi but its cool to see them keep improving the board.
All I use them for is random things like emulation of old games consoles…last project I did was to bung a pi zero inside a small 4 port USB hub (hard wired it to the hubs board) and it makes for a good 8/16 emulation machine
This one is a little bit more interesting as 4GB ram means there are options to use it as a lightweight PC and USB C power is a good idea.
I like the pi models on principle, but the processors just don’t have enough pep. The next project I have for the summer is prototyping model predictive contour control for an eventually autonomous formula electric car, and the pi should be a solid contender for a prototyping board but it just doesn’t have the umph for real machine learning. It’s a pity, because the rest of the hardware is good and the price is right…
Currently looking at the odroid xu4. Anybody have experience?
Is the differentiator needed? I’m thinking you could probably get by pretty comfortably with a PI controller for a water cooling system.
I swampt myself with extra projects right now, making a bag, building a custom hitbox, oxidizing screws, making a small carving knife and possibly moving country soonish if everything goes to plan, so I keep that on the back burner.
Probably going to upgrade my rig this year anyway, at least the cpu to a Ryzen 3800x + motherboard, so that might be a great opportunity for a bit of tinkering. You are right though, the differentiator part might not be needed but I would probably start with a full PID anyway and go from there. Right now I’m using a aquaero 5 pro which is a wonderful piece of kit, but something self build and a bit of tinkering on the side always excites me
I was having a chat with a chap on Saturday about machine learning with relation to autonomous vehicles; his company packages hardware (360 camera and lidar), the software, and a back-end service. Their client gives them ALL the data they gather using it, and they send back whatever processed results they get. They are now receiving more data than they can process.
If things go well, the odroid isn’t going to cut it either
Also when i asked him how he felt about the chances of things like security cameras and dashcams becoming increasingly cloud enabled and how excited he was about the possibility of getting near real-time data from multiple points and angles in any populated areas, he turned pale and suggested it’s best not to think about it.
Here’s hoping it get’s posted in the Google Stadia thread as another example of a killed off google side project!
Yeah, was kind of suspecting it. We’re offloading a lot of sensor processing and GPS communication onto seperate controllers where absolute state data is being produced, and the track is also being mapped on another controller…but still, machine learning is a CPU-cycle hungry motherfucker. Hm.
One benefit of it being a racing situation is that environmental awareness is considerably less detailed than what you need to know for a regular autonomous car on a road. Downside is that state estimation gets more complicated as you’re going fast enough that sensor drift is a serious concern, and the route needs to adapt to changing tire conditions, brake conditions, temperature, humidity…
I need a nap
They’re not doing any ML on their in field stuff so it’s not a concern, that’s all done on the back end. For both them and ourselves (we’re testing someone’s off the shelf solution) we’re running into memory being the issue.
I’d have thought a racing situation would be simpler because you have more known quantities and less variety than you do in general traffic, or is it specifically because you’re trying to do it in realtime mid-race?
I suspect the google answer would be to just send a million cars crashing round a track for however many years to gather enough data that their back end can use to determine what a perfect performance should entail, and then building a car specifically to do that, once, and then commence laurel-resting-upon.
Yeah, it’s the real-time issue. No back-end processing can be done, it all has to be situated on the car and adaptive. As such, memory is tight and processor bottlenecks are a problem to consider.
It’s simpler overall, because as you say there are far few inputs than in traffic. The main difficulty is that the goal is not just to drive, but to drive fast, and all while adapting the route to a bunch of physical concerns that aren’t as key at normal speeds. I don’t know for certain, but I would assume that stuff like brake heating, fuel tank weight load, unbalanced loss of tire tread, etc aren’t big factors in autonomous vehicles for the road.
The challenge, as I’m currently predicting it, is that rather than just making a route which ‘works’, the second you have one you want to adapt the internal coefficients to push a little harder, progressively. Keep trying for more speed. All with checks to ensure you don’t exceed traction limits according to sensor data from around the car.
That’s what the second million cars would be for!
There’s a battlebot competitor which it’s designer claims is partially automated so that it fires it’s weapon when something is in the optimal place to be hit by it. But it can’t aim, and therefore can’t account for motion of either bot. It’s hilarious to watch, it spends most fights flailing around on it’s side having missed the enemy and flipped itself.
Well someone was quick on the trigger , how you like it?
hehe - haven’t played yet.
I’m currently undertaking some training in new fields (I’m ditching years web design knowledge for pastures new) - getting a Pi was required for the next section of the course - so I figured I’d jump in and get a 4 whilst they were still in stock.
Can’t find my USB C power though… so I’ve had to order a new one…