Only problem with this is that mechanical problems are harder to fix at competition. At competition it is much quicker to fix a sensor or fix a peice of code than it is a mechanical part. Though if I may ask what sensor have the mechanical parts replaced?
These with the polarizing filter are very tough to fool.
This is a textbook bang-bang velocity controller with deadzone - definitely a closed loop system. What I came here to say is that while any control system has to control a setup that does indeed have hundreds, nay, thousands of variables, the magic of a closed-loop controller is that it will compensate for either not knowing them or knowing them badly - your classic open-loop controller assumes it has perfect information about its system, and that the system itself is perfectly modeled by your controller; it has to, since it has no feedback from the system it is controlling. The trick is, in order to actually get a perfectly modeled system, you DO have to consider all those pesky variables, or just tune it for a particular circumstance and pray it never leaves that island of stability. A closed loop controller can have a much rougher model of the system its controlling, and will magically find parameters that optimize its model (some of the really neat ones can even change their models to better adjust for different control regimes). Of course, you do have to worry about things like lag and complexity in your controllers, as well as having to sense things about the system you’re controlling to provide feedback, but thanks to the better modeling abilities of a closed-loop controller, it can be operated much closer to instability, which increases the rate at which it converges to the desired output. In short - if you need speed and precision, go closed-loop.
Our team has used many different sensors and controllers over the years. My favorite still is the organic analog computer coupled with a organic analog visions processor. Each year we put 2 of them behind the driver station physical connected to the driver station computer. Being an analog system the programming is very difficult. Training a neural net is very time consuming. This year we have devoted allot of resource to train our analog system. We went to ever off season competition we could. Got to train those nets.
I have no idea what you’re talking about, but the words ‘neural net’ have piqued my interests… You mean that you have a neural net managing your vision tracking, and that you have to endlessly test to train it?
I think he’s referring two a couple of students.
whoosh
You have no idea how relieved I am 
I thought that FIRST programming had just sailed over my head and continued to evolve while I wasn’t looking. My team was considering using an ANN last year, but we figured that the costs outweighed the benefits.
The SHARP IR sensors can be GREAT if used in the right application, but here are my lessons learned.
The version linked above is affected by long wire runs and supply voltage fluctuations. You can remedy this by putting a capacitor ‘backpack’ on the sensor, across the supply voltage pins (VIN, GND). I think we used soemthing around 300uF electrolytic cap. This significantly reduced our intermittent detections due to supply voltage fluctuation.
There are quite a few different models of these sensors available, they are each tuned to work at different distances. Make sure you get the right sensor for the job. You need to know the minimum and maximum distances the object you are trying to detect will be from the sensor. This page has a good break down of ranges supported by different models.
The larger versions that are mounted within their own plastic housing MUST be mounted so that the plastic is isolated from the chassis. I know it sounds crazy, but the plastic shroud on these is conductive.
Huh, we have spent a lot of time at competitions trying to get electronic stuff working, and failed. Mechanicals are easy to fix…maybe because I’m a mechanical engineer, with 30+ yrs experience fixing stuff. I guess it’s not what it’s made of, it’s what you know?
Our Ultimate Ascent robot was designed to not require sensors, it just does it’s thing based on it’s design. Autonomous just requires the robot to be placed where it belongs, and it shoots into the goal. Shooting during a match works by driving the robot into position and shooting…position being determined by parking the robot against the pyramid. Angles and positions are fixed by design, and the actuation is pneumatic, it’s either up or down, no “in between” positions that would require sensor feedback to control.
We’ve spend days, make that weeks, trying to get sensors to deliver useful information to control robots over the years, and generally failed miserably at it.