| |
In lugnet.org.ca.rtltoronto, Wayne Young wrote:
| |
In lugnet.org.ca.rtltoronto, Calum Tsang wrote:
| |
In lugnet.org.ca.rtltoronto, Wayne Young wrote:
| |
| |
Can I ask a stupid question from a non-moving bot owner? Why do you need
that much precision? The light readings are going to be pretty variable,
and the robot targets and obstacles are all moving, so do you really need
that much precision to know the general direction you need to be going in?
Calum
|
Are you thinking the robot could adjust its direction on the fly? In my
limited experience, precision matters when the robot is close to the TO,
especially if it can cover a lot of ground in the time it takes for the
sensors to come back with a reading (or to average a series of readings in
my case).
|
But if you are “close” to the TO...then your precision usually is higher
anyways--less movement, less slop?
Here’s some made up numbers to illustrate what I don’t understand...
If I read at time 0 at 30 degrees, a “contact”. By the time I come back to
it, at time 20, the contact is likely at 60 degrees. Why should it matter
that when I return at time 20, that I’m pointed to 31, 34 or for that matter
40? The target’s already at 60 degrees--so far away, any precision is
useless?
|
But suppose at time 20 your robot is pointed at between 10 and 30 degrees due
to the steering defect. Then it has lost even more ground to the moving
target.
| |
Obviously the next step is to scan again and try and reacquire the
light...every time I get scan, I should be getting closer, the final “steps”,
my turning accuracy should be higher if only for the fact I don’t “lose” as
much as a large turn. No need for high accuracy on large scans (eg, 360’s)?
Am I smoking crack here?
|
Your thinking is probably correct if the robot can sample the light reasonably
quickly, allowing it to rescan more and more often (and effectively take
smaller and smaller steps) as it gets closer to the TO.
Suppose the TO is stationary at 12 o’clock (or 0 degrees) relative to the
robot. If the robot can drive straight, then the path to the TO is a straight
line. Let’s say the robot has a defect that causes it to steer in the same
direction away from its target every time it starts moving after a rescan. In
this case the path is a series of zigzags. If the sample time is small, then
the zigzags look more like a curve, and the robot will end up at the TO. If
the sample time is really large, then the robot will end up on one side of the
TO. The extreme case is where the sample time is greater than the time it
takes the robot to cover the distance from its starting point to the plane the
TO is on. If the robot was really close to the TO relative to the TO’s width,
then that’s ok, the block transfer mechanism will probably still be lined up
with the hole.
You’re probably thinking that it’s ridiculous for the sample time to be so
large, and I’m thinking that I need to see if I can make my ‘bot read the
light faster so it can make faster decisions while on the move.
But yeah, castors are evil.
|
One reason the sampling time can be high is a misguided attempt to “smooth” the
times.
I tried averaging out the last five values in a shift register and found it
really slowed the thing down (my samples got really “steppy”). I switched to
just summing them without normalizing, which was faster, using a pointer to walk
the variables rather than copy them (which was faster) and then just switched to
a system of using newvalue=(oldvalue+sensor)/2 to give the current reading a
50% weight. This was the fastest and gave satisfactory results.
BTW: I hav abandoned my bot due to differential drive gear lash and castors...a
small memorial gallery is here:
http://ffaat.pointclark.net/gallery/view_album.php?set_albumName=RTL20
and have rebuild a new A/B skid steer chassis.
|
|
Message has 1 Reply:
 | | Re: Gear lash....
|
| (...) If I may comment on the castor issue. 1. Never put rubber tires on casters. 2. Use that funny boat hull piece as a caster whenever possible 3. Never put rubber tires on casters. The hole issue with casters messing up your direction control (...) (19 years ago, 3-Mar-06, to lugnet.org.ca.rtltoronto, FTX)
|
Message is in Reply To:
| | Re: Gear lash....
|
| (...) But suppose at time 20 your robot is pointed at between 10 and 30 degrees due to the steering defect. Then it has lost even more ground to the moving target. (...) Your thinking is probably correct if the robot can sample the light reasonably (...) (19 years ago, 2-Mar-06, to lugnet.org.ca.rtltoronto)
|
33 Messages in This Thread:
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
