ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

Thanks for supplying the bag file; that always helps in trying to debug this sort of issue.

It looks like the robot is consistently underreporting the actual rotation. I watched the scans accumulate in the odom frame in rviz (as described here), and would estimate that it's underreporting by 5-10 degrees per rotation.

amcl's odometry model assumes zero-mean noise. It has a hard time correcting for systematic bias, which your robot seems to exhibit. If your robot is indeed always underreporting rotation (more data might be required to verify that), then I would try to correct the odometry before sending it to amcl. You might try a variation on the TurtleBot calibration, which uses a wall as a fixed feature to compare to. Or you might just inflate the reported yaw differences by 1-3%.

Thanks for supplying the bag file; that always helps in trying to debug this sort of issue.

It looks like the robot is consistently underreporting the actual rotation. I watched the scans accumulate in the odom frame in rviz (as described here), and would estimate that it's underreporting by 5-10 degrees per rotation.

amcl's odometry model assumes zero-mean noise. It has a hard time correcting for systematic bias, which your robot seems to exhibit. If your robot is indeed always underreporting rotation (more data might be required to verify that), then I would try to correct the odometry before sending it to amcl. You might try a variation on the TurtleBot calibration, which uses a wall as a fixed feature to compare to. Or you might just inflate the reported yaw differences by 1-3%.

1-3%. Either way, if there's a consistent bias, it should be possible to do a one-time correction (not a periodic re-calibration). Such a fix could probably be incorporated into the YouBot odometry calculation somewhere.

If fixing the odometry is not feasible, I would try tuning amcl's odometry parameters. In this situation, odom_alpha1 is probably the most important one to increase, with odom_alpha4 coming next. I would leave the other parameters low (at their defaults).

I would not increase confidence on the laser model; the Kinect isn't really giving you high-quality laser scans. Instead, I would take the laser model configuration from turtlebot_navigation.