I also installed a new water cooled spindle on the router to play with, we’ll see if it is any significant improvement over the stock router. Just from spinning it up, it’s significantly quieter than the stock router. I’m going to be investigating ways to upgrade the rigidity of the machine in the future too, the shapeoko machines are severely held back by the belt drive system on the X and Y axis, and need to be improved to compete with similarly priced machines I’ve used like the Omio X8.
We will also be switching to Fusion 360 for our CAM this year, and will be teaching our students proper CAM techniques to help further improve our CNC work
I hope that y’all had a fantastic kickoff. There’s a lot to absorb and think about. Our team met from 8:30 this morning to 5:00 this evening, and we have some observations and have made a few early (Jell-o, not solid) decisions. As I convey these to you, please keep in mind the following guidelines.
We have team-specific goals that guide our decisions. These probably do not align with all the goals your team has made.
This is early stuff, we reserve the right to be wrong and change our minds.
We are particularly interested in Keeping It Simple while being highly competitive (within our team goals). We absolutely want to make a robot that doesn’t have unnecessary complexity.
The 2013, 2017 and 2019 games have something to teach us about cycling with unusually shaped game pieces (speaking specifically of the cone here), especially when they fall on the floor or are in a non-ideal orientation. This game has the longest run length for a cycle; longer distance than any of those three games. The game piece is probably most similar to the 2017 piece in terms of how annoying it will be if it’s dropped on the floor, like, in front of your loading area, for instance.
The “cube” really doesn’t need to be treated as such; it needs to be treated as a pick-and-place ball, like a beach ball version of 2019. The 2018 cube-orienting jaws are way overkill and unnecessary.
L1, L2 and L3 scoring are not equivalent to the 2019 game’s. L1 is on the floor, where you can literally score points by missing your L2 or L3 target. Height is only advantageous in the sense that you could pack more horizontal reach into a taller package.
Maximize possibility of winning matches and gaining the other RP’s.
Move quickly across the field.
Balance (“ENGAGE”) the CHARGE STATION in both auto and end game to improve both raw score, and the chance for an RP.
Collect cones from the floor in any orientation.
Place cones on the highest level, as early as possible, then move on to the next level.
Automate scoring as much as possible, using retro tape and April Tags to align and “just push the button”.
Floor cone collector.
Cone placement device.
extensive use of gyros and accelerometers to allow balancing in unpredictable circumstances.
small footprint, to allow for more balanced partners.
Cube collector is a lower priority for early prototyping; can likely be added to cone placement device later.
We can see a viable path for many teams to make very simple robots that play important roles in this game; there are a few more archetypes than we typically foresee. Cubes will pop all the time. Also, somehow I won $5 today from a bet I made with my team back in September.
We have learned a few things, and rethought a few things.
IS THE CONE A 2017 GEAR?
No. I was concerned that it would clog loading and scoring areas just by being against the wall, but as Spectrum clearly demonstrated here (and as we discovered as well), it squishes very flat. Right now it feels more like an oddly shaped ball that might require an extra stage to line it up for scoring, but that can use intakes that are fairly standard.
CAN A 2022 TELESCOPING CLIMBER BE USED AS A FAKE PINK ARM?
Yes, but if we end up using this we will extensively modify it. The bearings, internal friction blocks, and CF springs are all not ideal, and the tubing is very heavy considering how much of a moment arm it will need to produce. We tested one of our existing climbers turned on it’s side at 90, 45, and 60 degrees relative to vertical. At 45 it bound up a lot, less so the closer we got to horizontal. Here are two videos we took of the testing.
Another small update from our prototypes and testing today. Inspired by some of the testing team 3847 has been doing, we wanted to see what kind of compression we can put on the game pieces without popping or permanently damaging them. We built a simple test frame from some 2x4s we had, and drilled holes every inch to adjust compression levels. We used compliant wheels in our testing today, we did not test with more rigid wheels. We found that we were able to compress the pieces a lot more than we thought. At every compression level, both the cone and the cube fed through the wheels and were not damaged in any way. The only exception to this was the final test we did with the two rollers almost touching each other, we only tested that level of compression on the cone. Some of the videos at max compression are below because I think they are funny to watch. If you want more specific compression tests let me know and I can post videos of our results. Between the testing that team 3847 has done, and this testing tonight I’m confident that the game pieces are more likely to be popped or damaged by things such as sharp corners or screw heads than by compressing them against a wall or in a claw. Good news after how easy game pieces have been destroyed in the past few games!
For our first update working with vision, we spent some time getting Photonvision running on an Orange Pi and testing the performance with Apriltags. When we first got it working, the Orange PI was running 3D pose estimation for Apriltags with fps under 10, and latency above 15 ms on the lowest resolution available. In this example we connected directly to the Pi via Ethernet. We are also using an Arducam.
The next day, it started working at 100 fps with latency sometimes lower than 5ms, also on the lowest resolution available and connected directly. The camera was a little toasty, so we added some more cooling to the camera and the Pi.
On the lowest resolution, which was about 320 x 240, we could detect the tags about 10 feet away reliably. This could possibly be increased a bit with better tuned settings, but with high fps and very low latency, it is a bit overkill to have the resolution that low. On the maximum resolution, 1280 x 800, we were getting ok fps around 10, with latency around 50ms. With the resolution a step up from the lowest, at 640 x 480, we were getting 50+ fps with under 15ms latency. Performance with the Orange Pi is great, and it should be more than enough for doing vision this year.
We are done with our meeting times until Saturday, and it’s been a productive week. Very excited to take a few days off from things and absorb ideas for a while. Also excited to sleep and spend time on other stuff!
Today we CAD’ed and CNC’d several intaking prototypes. Since we are Open Alliance we will be happy to share these files with anyone interested; here is the link to the intake roller you’ll see in the videos below.
Prior to this, we tested the “pinch” spacing (less than 1/2") in order to see if we could make the cube and cone behave anything like the robot shown here (Iron Panthers 2018) and the answer was no. The medical wrap we used for this test (wrapped around 1.25" polycarbonate tube rollers, spares from our 2022 robot) literally eats the cone material when squeezed too tightly, and the cube just gets rejected after a quick expansion. As shown above, this is about 2" of space between the rollers (3.25" center to center), which worked much better with both game pieces. We were able to get the cone to align itself in the rollers from a number of orientations. We believe that with the addition of another roller and some height adjustments, we might be able to intake both cubes and cones from the floor, and align cones vertically from most angles of attack. This would be a game-changer for us. We will work on this design, create a new prototype, and test it on a moving chassis on Saturday.
NARROWING THINGS DOWN
It’s going to be incredibly important that we be able to decide on a design and construct it with time to practice cycles and work on programming. We are looking at several factors that will influence our final decisions.
Robot Architecture. We want to have as few degrees of freedom as we can manage. We want to reduce drastically the risk of a ridiculously long swing arm and the ensuing carnage (with misalignment being a major problem). A crane (influenced by the robots of the 2013 BEST Competition) and an internal passive alignment (shown here in our 2016 off-season project) are the leading contenders for making a simple robot that can score quickly from all positions. We will let you know what we go with (maybe both).
Game Piece(s) Manipulator. This may actually be less important, because we think whatever we choose to go with it will be modular and replaceable with something better later on. Currently the leading candidates are a full-width horizontal roller intake (shown above) or a vertical rolly claw.
Vision. Accurately aligning for both intaking and scoring game pieces will be critical and very difficult this year. Teams that think their success in the 2022 game will transfer easily to 2023 are badly deceived and need to face the fact that both acquisition and scoring are much more difficult this year, even discounting that the game pieces are not familiar and behave differently to the balls we’ve had to play with since our seniors were in 8th grade. Be wary! It does look like using vision is much easier if you are further from your goals, rather than parked right in front of them (where the April Tags are). Also worth noting that retro targets disappear once game pieces are scored on them.
That’s all for now. Be safe, have fun, and take breaks!
Did you try your test with the arm rotated 90 degrees to see if it made a difference? Thought is the weight is pinching the spring, if you rotate the arm does it relieve it enough to not bind? No idea if it would help!
Yes, we tested it vertical, horizontal, and at different angles between. It would take quite a few modifications before i would feel confident it would work consistently, due to the binding you referenced.
Did the cones seem to permanently deform at all after going through the squeeze? I am nervous if intakes like that catch on, that over the course of the season you are getting some gnarly shaped cones.