48128Re: [SeattleRobotics] Predicting Robot Position in a Differentially Steered R...
- Nov 13 5:27 PMLuke,I did this about 15 years ago but I used a cheap caster with a fat plastic wheel and changed the wheel out for a thin wheel with an encoder mounted on the side. Because of the problem of twisting the encoder leads, I put two stops on the bearing part of the caster to only allow it to travel a bit over 180 degrees of travel. It flopped around a bit on turns but seemed to accurately track distance.Good luck,Tom
I have been thinking about putting a similar encoder mechanism on my hacked roomba bot, but I decided against it because I was worried about twisting the cord that interfaces from the controller to the main control board. I assume that since the caster can rotate 360, there's a chance that the cord will twist itself beyond usability.
I would like to see how you built your caster/encoder assembly. Is there any way you could upload a photo somewhere and post the link?
this problem has stopped me from working on the autonomous mode of my robot for going on a couple years now. It would be great to find a working solution.
---In SeattleRobotics@yahoogroups.com, <twcarroll@...> wrote:Matthew and Robert,I've been following your inputs to the SRS site concerning encoders and problems with odometry. One method that I have used to increase encoder input accuracy is to use a separate encoder on a single trailing wheel that swivels much like a caster. If it is placed behind the robot in the center or even beneath the robot between the two differentially-driven wheels, the output 'ticks' will give you an accurate representation of the distance traveled.This single encoder and wheel is neither driven (no skids here) nor does the encoder array cause the wheel to drag, so it accurately represents distance traveled over smooth surfaces, short or long nap carpet or even uneven outdoor terrain. Unlike driven wheels that can skid and causing the encoder to read this as 'extra' distance traveled, the free-wheeling caster just traverses the ground or floor and swings back and forth to cover the robot's path. However, the 'distance traveled' data is an average of both driving wheels and is not used in turns where one wheel is driven faster than the other. When used with a MEMS gyro and / or a compass to accurately measure degrees of turning, this method can result in accurate odometry in complex paths.Good Luck,Tom C.
Subject: Re: [SeattleRobotics] Predicting Robot Position in a Differentially Steered Robot While TurningI've listened to and read about many people dealing with this and similar issues but it has never interested me, as a model. I mean, pick it up, the wheels spin, and it thinks its where its supposed to be. I prefer the idea of using gyros for motion sensing and/or, if you can, a laser range finder. For gyros, you can use an android phone... and use a USB to DB25 connector for I/O lines.For laser range finder, you can buy an expensive rotating one or buy one for hunting and rotate it yourself on a stepper motor. The algorithm is simple: For every point on the map, calculate how many points around you match in distance to what it would to surrounding objects, if you were there. You are calculating the probability of being at each point on the map. Then take into account where you were previously + the distance you think you moved. Do this between each point. It's very easy to narrow down you location, very quickly. This is how Google cars do it.
- << Previous post in topic Next post in topic >>