pic: 3CIM Ball Shifter

[AdamHeard]

Originally Posted by apalrd

Our driving theory around low (this was not always the case, but several of us on 33 have converged on this point independently) is that accelerating from a standstil is better in low gear, usually to around 60-80% of the low gear speed, than in high gear. So any time where we never reach peak speed in low, it's faster to drive in low. These are marginal gains in time. BUT, since high gear is in the 'bad side' of the motor power curve for the entire time (with 2:1 or more spread), the current draw from high gear will be WAY higher than in low. For the same or worse dynamic performance. This becomes MUCH worse on a dieing battery, which can frequently be seen near the end of a match.

We've sometimes run auto-upshift software to automate this, it begins the shift around 60% of low gear speed, but the shift takes time to execute under load (not sure exactly what speed it's at by the time the dog disengages).

We also run auton in low gear usually, for precision/control reasons, so a slightly high low gear is good for this. But we could run in high, we just don't, so it's not a huge objective.


I guess this differs from the 'west coast' opinion that 'high gear is where we operate, and low gear is just in case we need to push'. We operate in both gears.

I don't disagree with your points at all, I like the level of automation you've built in.

I don't know that anyone has run a decent autoshift out here.

I can certainly appreciate the lesser current draw, a dirty secret of some teams out here is we buy a new set of batteries every year, and if we didn't we'd like show worse performance on the field.

Part of this is we practice A LOT, so we use the last 1-2 seasons of batteries for practice to keep the new seasons worth nice.

[Andrew Schreiber]

Originally Posted by AdamHeard

I don't know that anyone has run a decent autoshift out here.

33 Ran an autoshift in '04 with their 4 speed. The the whitepaper is http://www.chiefdelphi.com/media/papers/1580 but it doesn't go into much detail on the shift algorithm. I'm sure Jim has/would describe it somewhere.

[AdamHeard]

Originally Posted by Andrew Schreiber

33 Ran an autoshift in '04 with their 4 speed. The the whitepaper is http://www.chiefdelphi.com/media/papers/1580 but it doesn't go into much detail on the shift algorithm. I'm sure Jim has/would describe it somewhere.

They're not out here

[apalrd]

Originally Posted by Andrew Schreiber

33 Ran an autoshift in '04 with their 4 speed. The the whitepaper is http://www.chiefdelphi.com/media/papers/1580 but it doesn't go into much detail on the shift algorithm. I'm sure Jim has/would describe it somewhere.

Our 11 code has the full implementation of it (for a 2-speed), but the code is a bit messy and I would implement it differently now. It's drivetrain/autotrans.vi. Upshift was enabled (and the driver feedback was all good), both downshifts were disabled due to driver feedback (Specifically he would rely on coasting into the peg with a tube, and the coast down would throw off his maneuvering as he was slowing down). Better calibration could have likely improved this.

The logic is fairly simple. We do three types of shifts:
Upshifts
Coast down shifts
Kick down shifts

Acceleration should be multiplied by the sign of velocity (or the abs of velocity should be used to calculate acceleration) to normalize for changes in direction.

Upshifts are based on thresholds for speed (greater than), abs of avg of throttles (for skid steer - this is pre-halo and culver drives) (greater than), vehicle acceleration (greater than), abs of diff of throttles (less than) (not turning)

Coast down shifts are based on absolute low speed (~2fps).

Kick downs are based on speed (we used 8 fps, which is higher than redline in low) (less than), abs of avg of throttles (greater than), not turning, vehicle acceleration negative and less than calibration (large negative number).

Upshift handles normal upshift driving.
Coast down shifts back when the vehicle speed is close to zero so it can upshift again at he next launch.
Kick down shifts down when you hit something and need to push.

This algorithm worked well enough for FRC. The corner cases (when turning) are simply ignored by the autoshifter, which was 'good enough'. We also impose a minimum time between shifts to prevent gear hunting, most automatic shift implementations see this and our solution is a 500ms inhibit timer.

[magnets]

How did you stop kick down shifting from happening when you ran into stuff? I tried a similar thing, but it would wear the shifting dogs out pretty quickly when we would hit something that couldn't be pushed.

[apalrd]

Originally Posted by magnets

How did you stop kick down shifting from happening when you ran into stuff? I tried a similar thing, but it would wear the shifting dogs out pretty quickly when we would hit something that couldn't be pushed.

The whole purpose of the kick down case is to shift when you hit stuff.

The dog is a fairly robust part, have you actually worn one (of the AM stainless dogs) out?

[magnets]

Originally Posted by apalrd

The whole purpose of the kick down case is to shift when you hit stuff.

The dog is a fairly robust part, have you actually worn one (of the AM stainless dogs) out?

Yup. We managed to break one in half with our auto shifting code several years ago. We had made our own transmission with two CIMs + another motor that was based off of the AM shifting transmission. The problem was that when our driver would hit the walls in the room we were testing, it would shift into the slower gear violently. Now that I think back to it, the problem probably occurred because we were testing with a large diameter cylinder because our little ones didn't arrive yet, so we increased the pressure to 120 psi to get a faster response when we shifted.