So over the past twoish years (summer of 2023 and all of 2024) I’ve been trying to figure out how to make effective autos that work 99.9% of the time and have been struggling greatly. Basically, I’ve never managed to make an auto that:
does more than just scoring preload and leave
works basically every time
In 2023 main season I did manage to make such an auto and I think that’s part of the reason why we did as well as we did. It was simple to just work and smart enough to always end in the right spot (balanced on the charge station)
it just did this:
score preload
drive all the way over the charge station (to get leaving points)
drive back on to the charge station (just enough to make the robot tilted)
use PID and the gyro to balance fully
and with those 4 steps we had an effective auto that with tuning worked 99.9% of the time.
Now come to the 2023 offseason and 2024 season we’ve invested in swerve and I’ve been trying to go for the extra pieces on the ground and I have been struggling to actually make the robot do anything consistently even if said thing doesn’t even get another piece. At our offseason competitions I have made some progress because of lots of time and tuning but its still failing to work even 50% of the time and even when it does work, it doesn’t contribute as much to the match as our auto in 2023 did.
So I have these questions,
How do you make effective autos with swerve that work 99.9% of the time?
Can I make these autos as simple as I had them in 2023?
How do you effectively tune swerve so that it is capable of running said autos?
If any of you have some other advice on this please do tell. I may also revive this thread in a month to ask about doing auto in 2025 but for now I would like to just use 2024’s game.
Now setting aside the 99.9% thing as hyperbole, the teams that do well design the bot for auto (Maximum envelopes for mechanisms to work), try to understand all the little details, practice a LOT.
That is up to you, there is a cost benefit equation/ tradeoff here. You will likely want to have a conversion with design/build about what is possible and in what timeframe. In general : simpler robot is faster to build and more robost therefore more time to work on autos. This obviously isn’t a linear relationship, but don’t go building a 4 DoF arm and asume it will be smooth sailing, that’s a one way ticket to sure failing.
A subset (successful autos) will shrink as you add conditions (more complexity)
Question 3 can be answered by looking at question/answer one.
Testing. A lot of times the autonomous is something that gets added in very late to the project. The way to insure reliable software is to test it, lots of iterations, vary the conditions.
Most of autonomous is movement. Something your can get started with via simulation, if you have a spare chassis that’s an excellent way to start working . (The latter is an expensive option)
But time and testing is the only way you get reliability up.
Welcome to a never ending rabbit hole, enjoy your stay.
Thinking of crescendo, the big goal was a 4 piece auto but that comes in pieces. You can try to first get two pieces until that’s perfect and then build off that. Also make sure it works on both sides (personal issue). If you have the ability, then you can keep going to the full 4 piece or greater however some can be limited by robot design. If you have a pass off mechanism time is wasted during the pass off.
The alternate is an approach outlined by 111/112 build thread for piece detection and always hitting it. Then if you can make that perfect it is point and click with some tweaking. Start slow and slowly build up speed and optimizing movements.
TLDR, start small with confidence and achievable goals and then start ramping. Design matters so consider that early in the year. Best advise is calibrate 3d pose with April tags and use the resources available (advantage scope, path planner, photon vision etc)
Looking at top teams in our champs division, I definitely noticed that many of them were running the same auto routine 80-90%+ during their prior competitions. I thought I might see more variety or even other wrinkles. Certainly those teams have those capabiliites. But the best auto is one that you’ve practiced and done many times.
I was judging FLL and asked the team about how they know their robot routine will work, and they said they said they use the Kevin Rule. That if they do it 7 times and it works, that it will probably work in competition. I also learned that Kevin was one of the kids on their team, and inventor of the rule.
Really want to underscore this, the mechanical design team needs to design for autonomous too, it is not just a programmer challenge.
In the post you compare the 2023 charge station balance to picking up game pieces off the floor in 2023 and 2024 and then scoring them, and this is a pretty big spike in difficulty.
Picking up a game piece in auto is greatly eased and accelerated by having a wide, not-picky intake.
Scoring after picking up a piece is trickier, you either need very robust finely tuned odometry (so the robot can drive to the exact right spot for scoring without drifting), or some sort of apriltag detection to correct for this random drift.
Cameras are your friend. No matter how good your drive team is aligning the robot, even the slightest error will be exponentially worse at a distance.
I always use the example of how much left/right error do you have if you were off 1* and trying to get to the center line 27’ feet away. The math is 27’*sin(1), which works out to 6”! That has been half a cube/note width.
We switched to a quality swerve/odometry library (YAGSL) and AprilTag vision, and we ended up always touching the close notes. The libraries have done great work for positioning.
I honestly think teams have been too reliant on odometry, and need to do more camera piece detection. How many times did you see this season teams go for notes that were missing in auto? They just follow their path, not adapt to the situation.
(Now I’m probably being hypocritical saying what you need, our autos have struggled recently, but have been improving. You really need lots of development time to perfect autos, and we finished mechanically 48 hours before our first competition, which meant the autos struggled)
Edit: I forgot to add, remember FIRST fields move during the event, expect them to move 1-2” throughout the event from all the impacts. Another reason why vision is important.