Hi Jolin and Bence,

I have found that using a long range IR sensor works better than sonar
when trying to substitute for a laser scanner in the ROS navigation
stack.  You can see my results here:

http://www.pirobot.org/blog/0014/

Most of the hard work behind these results was done by Mike Ferguson
from Vanadium Labs who makes the ArbotiX microcontroller.  We used the
ArbotiX with an AX-12+ servo and a Sharp GP2Y0A02YK IR sensor (range 0.2
meters - 1.5 meters).  Mike wrote firmware routines for the ArbotiX to
get 30 IR readings per 1 second sweep of the servo.  He then wrote a ROS
wrapper around the whole thing so we could use it with the navigation
stack.  You can see our whole discussion on the project at:

http://forums.trossenrobotics.com/showthread.php?t=4410

Note that I have only used this setup for obstacle avoidance and
navigation by dead reckoning via RViz.  Mike has done more extensive
tests with gmapping and his conclusion so far is that the panning IR
just doesn't give enough data to get consistent results.  However, you
can still go a long way with obstacle avoidance and path planning even
without full SLAM.

Patrick Goebel
Behavioral Sciences
Stanford University