Line Tracking Times?

[indieFan]

For those of you that successfully used line tracking last year, how long did it take for you to get your robot from the starting point to the wall?

Thanks,
indieFan

[Astronouth7303]

Last year's times would be irrrelevent, last years rc was WAY slower. So if you went a decent speed, You would over shoot the line.

[Cory]

It took us 15 seconds. I dont think anyone else did it faster (As always, correct me if Im wrong). The previous poster is right about the RC being too slow, at least to some degree. I think that at high speeds, you will still lose the line. It will be much better than last year, however.

Cory

[Greg Ross]

I already posted that our line tracking "worked", so I guess I had better chime in. But, as I posted before, we never used it in competition, and I don't think we ever timed it. I just know that it was way slower than the "blind swoop" we ended up using. (Probably someting like 6 seconds for the blind swoop, and, oh, maybe 12 seconds to follow the line?)

[Sachiel7]

You mean...
You mean people actually did get to the wall before auto mode was shut off?
Our linetracking was way too slow. The new controller isn't going to help as much as to make a big difference in the total speed of the mode. Since this years line is mostly straight, however, if your drive doesn't arc, it should go somewhat faster than last years.
This year, we aren't using any sensory in auto mode, except for the current sensors, possibly, and a few small tidbits. We found last year that running a pre-programmed (or pre-driven) path without any error checking worked just fine, and this year it should work even better, since bots have a much lower chance of colliding before their destination.
Anyway, if you're going with linetracking, the code is very simple, and the sensory required isn't that much. If you want to improve your speed, here's a few tips:
*Use 3 banners, Left - Center - Right.
*Make sure your bot is almost always going forward. Here's a movement table:
LEFT. CENT. RGHT.
LEFT HL LL F
CENT LL F LR
RGHT F LR HR

Where: H = Hard, L = Light, R = Right, L = Left, and F = Forward

When no signal is present, I suggest you add in a small "buffer" into your program to look at the previous few seconds of input. If it sees that it had a signal on the right, and lost all signals, then it kows it needs to make a sharp sharp right. In all the movements, even hard turns, include a bit (25% or so at least) of forward motion. The only time you shouldn't go forward is when you lose the whole signal.
* Other sensory. The more sensory, the more data your bot has to compare against other data. Don't be afraid to add in other sensors in-combo to help you get where you're going. For example, someone who really wanted to be accurate in there path could use both the IR and Banner sensors to traingulate a better sense of positioning on the field. Gyros can be useful in calculating the robots position as well.
Look around, and check out your options.
Anyway, those are just a few things I remember we did and it helped quite a bit.

[KevinB]

This year's line sensor bots should be quicker than last year, because this year the line is mostly straight with two abrupt corners, instead of being mostly curved. As such, once you get to a straight section, once your bot it headed the correct direction, it should be able to zoom on until it hits the next corner.

The more line sensors you use, the better you can optimize the program, lessening your chance of "losing" the line.

I would use all four banner sensors in the following configuration (all located as far forward on your bot as possible): one pointing straight down at the line (should start right above the line), one to the left and right of the line (should never see the line, unless the robot is "losing" position on the line), and one in the center pointing at the line a foot or so ahead.

By using the fourth sensor to look ahead, you can prepare for the two abrupt turns (and if you add a switch to let your robot know which side it's starting on .... you can program it to go ahead and start turining that direction when it sees the turn coming up).

P.S. With all that said .... I still prefer dead reckoning.

[Astronouth7303]

Originally Posted by Sachiel7

Anyway, if you're going with linetracking, the code is very simple, and the sensory required isn't that much. If you want to improve your speed, here's a few tips:
*Use 3 banners, Left - Center - Right.
*Make sure your bot is almost always going forward. Here's a movement table:
LEFT. CENT. RGHT.
LEFT HL LL F
CENT LL F LR
RGHT F LR HR

Where: H = Hard, L = Light, R = Right, L = Left, and F = Forward

When no signal is present, I suggest you add in a small "buffer" into your program to look at the previous few seconds of input. If it sees that it had a signal on the right, and lost all signals, then it kows it needs to make a sharp sharp right. In all the movements, even hard turns, include a bit (25% or so at least) of forward motion. The only time you shouldn't go forward is when you lose the whole signal.

Huh?

[Sachiel7]

Sorry for the 2 Ls in the table. Let me give you an updated one:
________LEFT CENTER RIGHT
LEFT_____HL____LiL____F
CENTER___LiL____HF___LiR
RIGHT_____F____LiR___HR

Where H = Hard, F = Forward, Li = Light, L = Left, and R = Right

This is a map of the sensor input. For example, when the Left and Center sensors are on, you move Lightly to the left. And, as I said, every bit of movement should have a certain percent of forward motion. So, when it goes lightly to the left, it really is going forward and left a little. The harder the turn, the less forward motion, but it should be there.
If you're still confused, let me know. These were just a few tricks we used last year to speed up our linetracking.

[WillyC]

Good call on the look-ahead sensor and the "what side of the field am I starting on?" switch, KevinB. We were also going to use the switch to let the robot know which side of the field it's starting on - left or right.

But we hadn't thought of the look-ahead sensor; thanks for the tip. Have you actually tried that look-ahead sensor? If the banner sends out a signal at a downward angle toward the front, it might not see the reflected signal. I tried this and found that 45 degrees from vertical is about the most you can angle it. Best to mount that look-ahead sensor out front, even in front of the two "straddling" sensors.

We only got two Banner sensors in our kits; did you order the other two? I was thinking I could snag two Banners from another team if they're not planning on using theirs...?

[KenWittlief]

we had line tracking working well last year - and it took us about 5 seconds to get to the wall with it

we had 7 banner sensors on a bar about 2 feet wide.

funny thing is, the way we got our line tracking to work, we actually lost the line (went past it) which left the bot in a hard turn, and it would reacquire the line and then shoot up the ramp.

but we never used it in the competition. After we shipped the bot we realized it would be faster to use the gyro sensor, which we already had installed for closed loop steering, to integrate a compass heading, and to start off backwards and do a V turn.

this got our time to wall down to 3 seconds.

BTW - there was something about last years line following that made it very difficult - the half circle connected to a straight line requires accelerated acceleration to follow - in physics its called a jerk

the discontinuity from a straight line to a circle looks smooth, but its not. In fact, to follow the line perfectly last year would have required infinite acelleration (impossible)


and guess what, those bends in the line this year? same thing - you cant follow them perfectly, so your algorythm will have to take that into account - anticipate where you are going (straight, left turn, right turn....)

[KevinB]

Originally Posted by WillyC

We only got two Banner sensors in our kits; did you order the other two? I was thinking I could snag two Banners from another team if they're not planning on using theirs...?

Oh ... for some reason I thought there were 4 Banners in the kit. Am I going crazy?

[WillyC]

Originally Posted by KenWittlief

BTW - there was something about last years line following that made it very difficult - the half circle connected to a straight line requires accelerated acceleration to follow - in physics its called a jerk

the discontinuity from a straight line to a circle looks smooth, but its not. In fact, to follow the line perfectly last year would have required infinite acelleration (impossible)


and guess what, those bends in the line this year? same thing - you cant follow them perfectly, so your algorythm will have to take that into account - anticipate where you are going (straight, left turn, right turn....)

Interesting point...

I don't think it's realistic (or at least necessary) to follow the line "perfectly". If you use two sensors, one on either side of the line, and space them out a bit (i.e. for a 2" line put the sensors about 4" apart), then the robot will "waddle" down the line. You know that when you get to the end of the line, the robot will be within 2" horizontally of it's target. It may be pointing at a bit of an angle, but it will be there. If you combine this with some other sensory info (turn counting, timing, gyros etc), you have a pretty robust system. But not perfect...

[Astronouth7303]

Actually, I know of a method that uses only 1. Program the bot so that:

Sensor
vv Motor
XXX Left
DRK Left
LIN Right
DRK Right
LIN Left
DRK Left
LIN Right
DRK Right
LIN Left
etc...

Basically, it waves over the line. However, 2 or 3 are far more efficient.

[lpramo55]

we set up a test for our algorithm with this years rc but diff. gear ratio and drive wheels than this years robot will be. the test included a 90 degree angle and several other twists....granted it was about half the size of what it will be in competition i feel that it is much more difficult and didnt allow us to reach our full potential speed so our estimate, in conclusion, is that we will get there in 5-7 seconds

[bob1million]

Originally Posted by WillyC

Interesting point...

I don't think it's realistic (or at least necessary) to follow the line "perfectly". If you use two sensors, one on either side of the line, and space them out a bit (i.e. for a 2" line put the sensors about 4" apart), then the robot will "waddle" down the line. You know that when you get to the end of the line, the robot will be within 2" horizontally of it's target. It may be pointing at a bit of an angle, but it will be there. If you combine this with some other sensory info (turn counting, timing, gyros etc), you have a pretty robust system. But not perfect...

Our team tried it with two sensors and we had the problem of over shooting the corners if we tried to have the robot moving at all times. If every time the sensor hit the line and we stopped and moved left/right motors it worked fine but it was kinda slow.