The Future Of VexAI

VEX V5 General Discussion General Forum
1

This last season has been a challenging year for all Vex teams. Nonetheless, we should look out for the future and what it holds. Covid-19 has been detrimental to VAIC's pilot year. Teams that had registered haven't been able to meet up as they could have in previous years. Still, that doesn't mean we haven't found ways around that.

In working with the new Jetson Nano-based system, we found many things we liked. The GPS sensors give reliable feedback to where the robot is on the field at any moment. The vision system even returns data on the exact coordinates and distance of a ball or goal.

We also found some issues. These issues lie not with the sensors we have but, with the ones we don't. As I take it, the point of VAIC is to build two robots that perform autonomously for two minutes. Unfortunately, the sensors provided aren't enough for the task of navigating a field with two other robots. Detecting other robots can be achieved with a distance sensor. This approach is simple, but it does not offer enough detail and information to be sufficient.

Vex AI teams need a way of seeing what is around them. We need a way to get information about the dynamic condition of the field. Getting data on what is around the robot is crucial to deciding how to move. Without this system, we can't advance our learning to where we want.

I'm not sure what is in the interest of Vex to allow 3rd party electronics. But for one, I enjoy going from the ground up and doing the things we usually don't get to do in VRC. I bring up 3rd party electronics because they form a wide range of possibilities for obstacle detection. I hear, saying this from rumors and speculations, that Vex will not allow 3rd party sensors next year. If true, that would mean there is little to offer for my problem. On the other hand, if it is possible to release such a sensor, that would be terrific.

To wrap this up, VAIC has tons of potential. Never have I seen a competition so easy to get into while also having plenty to learn. We simply need a way to see around. I look forward to seeing what comes next in future years.

Best Regards,

Trent

This is a personal opinion

3

We simply need a way to see around.

When our VAIC team first considered how to approach this game that exact question was raised. Without knowing about the goal states and robot positions of both alliances you’d quickly end up with unplanned or messy outcomes on the field. If you don’t crash into them they’ll crash into you!
Our approach to VAIC Change Up was to mimic the human visual experience, i.e. a person standing looking at the field, and pass that data to a single robot that performed scoring duties. 24" static robot with a mast and 15" scoring bot. Nothing went beyond prototyping however since there’s no possibility of our team attending an event this season.

The Change Up game design is tailored towards human players. VAIC would be much better served with a game tailored towards AI otherwise you’re always going to want to “see” and that drives the need for 3rd party solutions since the current focus of VEX AI sensors is only game object recognition and position. Disallowing 3rd party electronics and sensors would not be positive.

4

Seems like a good time to throw up some stuff from an old robot of mine.
You can see the value of a high FOV camera, 360 degree coverage of the field.

image
image2500Ă—444 1010 KB

image
image2500Ă—1875 2.1 MB

5

Can you post what camera you used?

I’ve poked some with the

https://www.robotshop.com/en/ydlidar-x2-360-laser-scanner.html

to try to distance sense for another project. For these games it’s going to also sense the field elements and there is going to be a blind spot. If you put it up at the top of the robot it’s much better.

But Tabor and my posts are saying the same thing, other sensors would be a must for success. (This is the @Trent point) Expecting VEX to come up with all the variety of sensors isn’t a sustainable business model for them. (Non recurring Engineering costs are a big deal) Allowing other sensors is important.

I’d like to see VexAI grow. To be honest, in my long past in IT, I’ve been burned by proprietary dongles to the tune of 10’s of thousands of dollars. So until that gets fixed, I’ll watch in awe what goes on.

But I for one, welcome our new Nano overlords. :wink:

6

https://www.amazon.com/dp/B00LQ854AG/ref=cm_sw_r_other_apa_i_8f.ZEbH1QWJ8K
Really just a generic USB fisheye camera.

That is effectively an updated version of the sensor I used here

High FOV is important for VAIC to work well, teams should be able to explore these different sensor options.

7

I only have a passing familiarity with it, but doesn’t FRC use retro-reflective tape on many of their field elements? Would requiring VexAI bots to have identification plates with retro-reflective tape on all 4 sides of the robot potentially help simplify the friend-or-foe issue?

Potentially more radically, could the field control software (and programming template) be updated to accept a heartbeat from each robot’s GPS and then broadcast that back out so all robots know where all other robots are at, say, 1s intervals?

Another alternative might be for VexAI to go down to 1-1 to help simplify friend-or-foe to “not-field element” detection…

8

FRC has really tried to make vision an easy thing, tape helps. VEX could consider IR emitting goals, but that would be an EP nightmare, I now regret typing it.

FRC wants your number on the bumpers, four sides, but I like the thinking that VEX could require reflective tape on 4 sides (a 3x3" square would help)

I agree, having the template put out GPS coordinates would be great. I know that serial connections between two VEXU robots is a thing, adding this would be a game changer. But harnessing my Dr Evil abilities, this allows a robot to zero in on you, pin you for 4 seconds back up 2’ and then re-pin again. (Another reason I’m not on the GDC)

9

Unless I misunderstand the manual, they are allowed in the AI competition.

image
image1440Ă—492 149 KB

10

This year that is true. Next year maybe not. The RECF has been on edge from what I’ve heard. I’ll give you a source in a second.

11

Good explanation, agreed.

12

Yes FRC uses retro-reflective tape which is very effective because your own light source is on-axis with your camera and thus it’s possible to essentially fade out most of the other parts of the image making image processing straight forward. Low cost computing moudles such as Raspberry Pi can handle it which is what LimeLight is based on. We used this last season with excellent results (until mid-March anyhow).

I’m surprised VEX didn’t opt for a retro-reflective bar code for their GPS, cost I guess.

13

IR emitting targets for autonomous period were used in FTC until 2015. What replaced them was worse and all FTC teams cursed the red/blue beacons until they were displaced by vision targets. Can confirm the IR emitters were a pain for AP’s to deal with. Much better if goals are passive identification.

14

@Jim_Crane has stated that the jetson and associated electronics bought for this year will be illegal next year, at least in VAIC-HS:

I’m not sure how to read this other than to say that all non-vex electronics will be banned next season for VAIC.

15

Thanks for the source.

16

You work under the assumption VEX Robotics will deliver a new product on time?

17

Even though we have not had the events we planned for VAIC, we have been moving forward with the advancement of this program, getting ready for the 2021-22 season, one season past the pilot season.

In the pilot season, we had 3rd party sensors that teams had to use because the VEX sensors were still in development and we needed to see what teams were able to accomplish.

As people have accurately pointed out, VEX U and VAIC allows for 3rd party, and this will continue to be true. But there is a HUGE however.

The software that we provide will be written to work on the VEX hardware. So, the GPS won’t work on 3rd party hardware and the AI won’t recognize game objects using the 3rd party hardware.

Teams that choose to use their own hardware will also need to build their own software. We simply don’t have time or staff to write software for multiple platforms and the documentation on how to use that software on multiple platforms.

So, I was wrong saying that it will be illegal next year because I knew it wouldn’t work and assumed we would make a rule so that no team was caught off-guard wondering why their equipment doesn’t work. But, in talking with the software engineers, there really is no reason for us to make it illegal. There are other things you might want to do with that equipment, and it might be interesting to see how you use it. Just know that creating your own GPS code and AI recognition software will take a team of developers.

That brings me to my next point. In this pilot year, we knew that we were not able to provide the level of documentation needed. Instead, we created a closed forum group where you could ask the VEX development team directly. But this is not effecient nor sustainable. With us moving to a single platform (only supporting a single platform), we are hard at work developing the training guides and documentation that teams need to be successful. There will be tons of knowledge based articles written to explain how to use the new VEX hardware. We aren’t ready for the roll out yet, but will be by early Fall.

So, you have changed my mind on making things illegal, but please be careful in how much you want to take on without any support or documentation as you navigate using 3rd party hardware. You will have a huge competitive advantage having the software supporting documentation and AI trained software when using VEX’s equipment. If you use last year’s, it will be very good at recognizing last year’s game elements, but none of this year’s will be known on your devices.

It was true that everyone was told and had to acknowledge in the application that the hardware you received would need to be replaced next season. This is no longer true by rule, but it is in spirit unless you are crazy good at writing for GPS and AI training.

18

Admittedly, I know very little about ROS, but it seems to be an open standard that allows various hardware/software to be incorporated in a robot using a pub/sub messaging programming model. Perhaps if Vex included direct support for it in the Vex API, teams would have the ability to use other ROS-compatible hardware/software without placing too much of a burden on the Vex programming staff in supporting non-Vex hardware.

Teams, then, would be able to potentially leverage the open-source world (with potential cross-over with the FIRST ecosystem), and may need not become “crazy good at writing GPS and AI training”

19

Ya I don’t post about it much but all my fully autonomous robots used ROS. This is exactly what it’s for, it let’s you focus on learning what you want to learn and supports you to use open source software.

Ris visualization tool showing state of my autonomous robot as it drives around map.

20

Honestly I´m confused about VEX AI and its target audience - if it is the college level it should be ROS with all the standard stuff like tensor models and open source etc That is what students should learn, and then using a the NANO and the Intel 3D cameras is like a must. (the two camera version NANO with openCV etc ROS etc is also a good option)

For advanced HS students the same goes - while that is a smaller subset, it is also an environment which in my opinion does not warrant nor justify proprietary tech, but rather openSource software and hardware and leverage what is going on in that space for learning. I would consider making the V5 brain an ROS node for controlling motors and some sensors interfacing with the NANO running all the hard stuff and the ROS master. Students learning ROS and neural nets / AI using tensorFlow, torchRT etc etc is really where the skill are going today. If an RPI can run a full ROS node/master and image process so can the V5 brain it has infinite more horse power…

In short I have a very hard time seeing where VEX AI is and is going at this point.

21

I think it’s a great concept, but agree it is hard to see what the target market is. What is the difference between VEXU and VEXAI? Driver control? Clearly a pair of robots for VEXAI could also do VEXU. But where does that leave high school teams? A venue for well-funded high performing teams? A transition to VEXU?

Vex already makes parts for FRC/FTC. I would think that making sensors that support ROS would appeal to FRC teams, but I don’t have much direct knowledge of FRC. From a business perspective having a line of products available to VEX*, FTC, and FRC would seem to make sense. I get that ROS might be tough for a middle school VRC team, and that VEX does seem to have a good “ramp” in their product space.

I see a lot of potential for VEXAI and would love to see it succeed. That said, the devils in the details.