RoboLoCo, Team 5338
Replies: 5 comments 6 replies
-
|
Does anybody know when we will be able to make purchases, and how long it will take for them to be actually put in? I want to know so I can begin planning a timeline for them. |
Beta Was this translation helpful? Give feedback.
-
|
I think you'll have to speak with Coach Ajima on that one. |
Beta Was this translation helpful? Give feedback.
-
|
Another thing to consider, especially in the case of the lower end processors, we need to consider how many processors we will need to purchase as well. This of course, goes hand in hand with #143, as this number will depend on how many cameras we want to support. |
Beta Was this translation helpful? Give feedback.
-
|
Raspberry pi 5 seems like a pointless purchase as they are close in price to the orange pi 5 pro. As for the coral board, I think it lacks the ram and overall processing power necessary for multiple parallel vision tasks. There is also the orange pi aipro, which offers a 20 TOPS accelerator with 8 gigs of ram for ~200 dollars. Between that, the jetson, and the mac mini comes down to what would be overkill for computer vision, and they all have downsides. Though the mac mini is very cost effective for its ai TOPS, its less suited towards CV and edge computing. The jetson has plenty of processing power for CV and probably can handle multiple cameras and tasks in parallel, but it is rather pricey and may be more than what is needed. The orange pi aipro does appear cost effective and is geared towards CV, but by what I have seen it has much less support and documentation for software. |
Beta Was this translation helpful? Give feedback.
-
|
This does bring up the concern of what budget we can afford to spend on this, especially considering that we ALSO need to buy cameras and potentially spare devices if possible. |
Beta Was this translation helpful? Give feedback.
-
|
I agree with a lot here. The Orange Pi 5 (any of its many variants), seems like a very good option. Especially compared to the Raspberry Pi 5. The Jetson doesn't really seem like a very good option in any regard. It's beat by price/performance both below and above by Orange Pi and Mac Mini, respectively. I'm also very much not a fan about how Nvidia has handled support for their Jetson boards. The Mac Mini is definitely incredibly powerful if the team has big ambitions on vision (or maybe other sensors like lidar?) processing. We'd likely only need one per bot but we'd have to factor in the poor support we'd be dealing with, as well as the size and form-factor to deal with on the robot. |
Beta Was this translation helpful? Give feedback.
-
|
I wasn't actually aware that the Coral TPU came in the form of a development board. I mentioned it as a possible accelerator on any existing dev boards with an M.2 slot (Orange Pi 5) or at least a spare USB port. The 2027 SystemCore will have an M.2 slot to be used with a similar accelerator card (Hailo-8, which seems to be in the $200-300 range compared to the $40 price for an M.2 Coral). But even if it's not supported directly, the 2027 SystemCore is just a raspberry pi 5 running a Linux distro so it's not impossible to say we could get it working there too. Just a fun idea though. Absolutely not necessary with the other options available. |
Beta Was this translation helpful? Give feedback.
-
|
To note, the Mac has many capabilities with Swift and existing Apple libraries that we could utilize to streamline our vision and autos. Major BenefitsObject Identification in Live Capturehttps://developer.apple.com/documentation/vision/recognizing-objects-in-live-capture Tracking Multiple Objectshttps://developer.apple.com/documentation/vision/vntrackobjectrequest/init(detectedobjectobservation:) Finding Trajectories of Flying Objectshttps://developer.apple.com/documentation/vision/detecting-moving-objects-in-a-video 3D Model Generation for the Fieldhttps://developer.apple.com/documentation/realitykit/realitykit-scene-understanding Minor BenefitsCoreMLhttps://developer.apple.com/documentation/coreml Recognizing Text from Imageshttps://developer.apple.com/documentation/vision/recognizing-text-in-images |
Beta Was this translation helpful? Give feedback.
-
|
Do you have a specific mac model in mind? What's the pricing? |
Beta Was this translation helpful? Give feedback.
-
|
Since all the current Mac minis that are selling have M4 chips, they should be good regardless of storage. Ram shouldn't be an issue considering we'll be only running pretty basic models locally, not training them. The base config of the cheapest mini is 500$. Apple charges 200$ per RAM upgrade, so that's why I'm reluctant on upgrading it as well. |
Beta Was this translation helpful? Give feedback.
-
|
FRC does have a price limit on individual components of $600 so we wouldn't be able to upgrade too much. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
If you don't already know, Vision's processing(for example, recognizing where an April Tag is and the angle to the April Tag) is not handled by the Roborio like other systems. Instead, we use another device referred to as a coprocessor to handle this processing. Currently, we are using a Raspberry Pi 4B for this. However, this is not a great vision processor, and there are many more favorable options on the market, especially if we wish to pursue Object Detection(see #114).
We need to choose what vision coprocessors to invest in for the next competition season. Ideally, we would have back ups for these coprocessors, however we may be limited by costs.
As for the options of what coprocessors we have, the obvious options are a Raspberry Pi 5B and an Orange Pi 5/5+. These are both better than the processors we currently have, and cost roughly $100. Of the two, I would choose an Orange Pi, because Photonvision releases Object Detection supported drivers ONLY for Orange Pi 5 or 5+s. It is important to note that either of these processors can only support 1 or 2 cameras before performance starts to take major drops.
Another well known option that was pioneered this year was the Mac Mini, which costs roughly $600. It is multiple times better than either Pi, and it can support many more cameras without taking a performance dip. However, if we want to use it, we would have to either swap to Northstar(which we are not guaranteed to be able to find resources on, given that it is intended for use primarily by Mechanical Advantage) or write our own driver.
However, these are not our only options. @VMFortress recommended one option of using the Google Coral. It costs roughly $130.
Another option me and @314PiGuy is the Nvidia Jetson Board`, which can cost up to 800$(but one link currently shows them on sale for 600$ IIRC).
Now, I'm not an expert in processors and hardware specs and stuff, so I'll leave the floor open to anyone who is more experienced in this area to make or eliminate suggestions. However, I would like to make this decision and make the purchase as soon as possible, not only to have this soon to test, but to also know what, if any, custom software we would need to write to support this.
In my opinion, regardless of what final coprocessor we choose, we should at least purchase 1 or more Orange Pi processors. This is because only the Orange Pi already has widely tested object detection software implemented with Photonvision(that I know of), and so provide a safe fallback option in case we find our other processor(s) unsatisfactory.
Beta Was this translation helpful? Give feedback.
All reactions