Thursday, May 27, 2010
Robot Perception Part 1
I have several ideas on how to carry out with this, I am first trying out IR Rangefinders. The IR Rangefinder is mounted on 2 servos, 1 rotating it side to side, the other up and down. The IR Rangefinder will scan an area and save it into an array of floats or doubles (depends on how accurate IR Rangefinders are) and then send that array to an external processor that will process that data. There will be simple trignometry done on that data (don't worry, I already got the algorithm) then return the X, Y and Z coordinate of the point. Then that vertex can be used to generate a 3D image or be processed even farther for robot logic. The major down side is the time it takes to get the data and the noise and it will not work for moving objects. That is a HUGE downside. I have several ideas using that basic principle: several of them working together, 1 scans on one axis all around the area and the other one can scan the area with an object detected in dept. Now those are some very basic ideas, but I still think that it will not be the best way to handle robot perception in a quick and fast dynamic world. Now there is another similar method of robot perception, but I will go into it in the next post...
Subscribe to:
Post Comments (Atom)
Using IR sensors seems very dicey for autonomous systems. If anything (robot, ball, etc) gets in the way of one of them for a fraction of a second during a game, then you are going to get inaccurate x/y coordinate data for the rest of the match.
ReplyDeleteWhy not try using encoders and a gyro? You can use math to estimate the distance traveled from the encoder and use the gyro to calculate over time the way the robot has turned.
I don't know if you have seen this yet, but it would be a great way to get started on this: http://www.wildstang.org/main/stangps.php
Thanks for your response, I saw that website, yea, I agree the IRs are dicey, I will post part 2 soon
ReplyDeleteIn response to the first comment: I don't think that gyros and encoders solve the same perception problem as range finders (or cameras or laser sensors). Traditional autonomous robotics requires two things: 1) Know where your robot is, and 2) Know where everything else is.
ReplyDeleteUsing a system like Stang ps is great when you are operating in an environment that is relatively static (like the 15 second auto period this year). However in a 2 minute match the field won't be static. It is sort of like the idea of trying to drive using only gps coordinates. On flat empty spaces, it works well, however in the real world you are going to smash your car into something.
I agree that a gyro and encoders (or another positioning system) are necessary, however they are not sufficient. By combining position data (from encoders.. etc) with perception data (from rangers or cameras) it is possible to both know where you are, and where what you are looking for is, leaving the robot to autonomously decide how to get from one point to the other.