Ask Your Question
3

Getting accurate real-time x,y coordinates with gmapping?

asked 2013-07-12 07:25:25 -0500

BlitherPants gravatar image

updated 2014-01-28 17:17:14 -0500

ngrennan gravatar image

Hello,

I have a Turtlebot 2 (Kobuki base) running in Ubuntu 12.04, ROS Groovy. I want to be able to print out a map of the robot's X,Y coordinates in real time.

Using information I've spliced together from various tutorials (still a newbie), I've come up with a node that uses a StampedTranform object to lookup the transform from "/map" to "base_link". It then takes the transform.getOrigin().x() and transform.getOrigin().y() and prints it to a text file.

My steps are this:

roslaunch turtlebot_bringup turtlebot.launch

roslaunch turtlebot_navigation gmapping_demo.launch

roslaunch turtlebot_teleop keyboard_teleop.launch

I then launch my application and manually drive the robot in a rectangular path around an office building (perhaps 50ft by 20ft). I load the resulting XY text file into MS Excel, and plot the points, but the path looks pretty awful.

I don't have enough karma to post pictures here, but suffice to say it looks like a squashed rhombus.

My question is, what am I doing wrong? Is there a different transform path I should be listening in on? Am I using the wrong function to extract X and Y? Or is it a bad idea in the first place to try and get the path coordinates in real time while the map is being built?

Thanks in advance! I've posted my code below, if anyone is interested.

#include "ros/ros.h"
#include "std_msgs/String.h"
#include <tf/transform_listener.h>
#include <iostream>
#include <fstream>


int main(int argc, char **argv)
{

  ros::init(argc, argv, "PoseUpdate");


  ros::NodeHandle n;


  ros::Publisher chatter_pub = n.advertise<std_msgs::String>("chatter", 1000);

  tf::TransformListener listener;

  ros::Rate rate(1.0);
  std::ofstream myfile;
  myfile.open("/tmp/OUTPUTXY.txt");

  while (n.ok())
  {
    tf::StampedTransform transform;
    try
    {
        //ROS_INFO("Attempting to read pose...");
        listener.lookupTransform("/map","/base_link",ros::Time(0), transform);

        ROS_INFO("Got a transform! x = %f, y = %f",transform.getOrigin().x(),transform.getOrigin().y());
        myfile << transform.getOrigin().x() << "," << transform.getOrigin().y() << "\n";
    }
    catch (tf::TransformException ex)
    {
        ROS_ERROR("Nope! %s", ex.what());
    } 


    rate.sleep();

  }
  myfile.close();
  ROS_ERROR("I DIED");
  return 0;
}
edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
3

answered 2013-07-12 07:40:32 -0500

tfoote gravatar image

If you're using a TurtleBot with the Kinect or Asus Xion a space that large is going to be close to impossible to map. The Kinect only has an effective range of 3m and with it's small field of view a lot of the time it will be out of visual range of any obstacle. And the odometry of the Kobuki is much better than the Create, but it's still not going to allow many meters of deadreakoning. And with the small laser scan width, it is much harder for gmapping to do good scan matching since much of the time it will see feature less walls in it's small field of view.

This is something that can be significantly improved but may require new tools and algorithms, not just tuning parameters. Or a new sensor such as a longer range/wider field of view laser scanner will help significantly.

edit flag offensive delete link more

Comments

It's a Kinect. The area I'm driving it around in has hallways that are about 5 feet wide (office with cubicles, etc). Is it really hopeless even with all the doorways and cubes in sight?

BlitherPants gravatar imageBlitherPants ( 2013-07-12 07:45:54 -0500 )edit

Also, if I drive around and do a map-saver to PGM, the output map looks pretty decent, so I had thought that the system was able to manage the office space...

BlitherPants gravatar imageBlitherPants ( 2013-07-12 07:50:32 -0500 )edit

It really depends on the space and how much you can keep the kinect scan line full. The office space we were testing in had 10' wide corridors. There are techniques such as doing perimeter driving which can help a lot. If you're getting a decent PGM of the map, you should be able to make it work.

tfoote gravatar imagetfoote ( 2013-07-12 07:53:33 -0500 )edit

By "perimeter driving", do you just mean hugging the wall?

BlitherPants gravatar imageBlitherPants ( 2013-07-12 08:02:21 -0500 )edit

Pretty much. Keep the kinect field of view as full as possible within 3m of distance.

tfoote gravatar imagetfoote ( 2013-07-12 08:06:42 -0500 )edit

That's encouraging to hear, thanks. Otherwise, do my methods of getting the transform look correct to you?

BlitherPants gravatar imageBlitherPants ( 2013-07-12 08:11:00 -0500 )edit

Ahh, sorry I didn't look closely at the code. I think you're asking for the inverse of the transform you want. Such that you're plotting the position of the origin of the map in relationship to the position of the robot. Try reversing the argument order.

tfoote gravatar imagetfoote ( 2013-07-12 08:20:44 -0500 )edit

You mean like this? listener.lookupTransform(/base_link","/map","ros::Time(0), transform);

BlitherPants gravatar imageBlitherPants ( 2013-07-12 08:25:12 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

2 followers

Stats

Asked: 2013-07-12 07:25:25 -0500

Seen: 3,856 times

Last updated: Jul 12 '13