How to set an estimated position without using rviz

edit

You could try using one of the lightweight rviz alternatives for machines with little graphics power. Those can be found here.

Additionally, you could create your own initialpose message and publish it to amcl that way (if you know the robot's pose). The initialpose message is of the type geometry_msgs/PoseWithCovarianceStamped.

EDIT: For this message type, the seq field is filled in by ROS automatically, and the stamp is zero, so you can ignore both of those fields. The frame_id is /base_link. Finally, the covariance is a 6x6 matrix that is filled in with set values. I've provided an example message below that was taken straight from Rviz.

header: 
  seq: 14
  stamp: 
    secs: 0
    nsecs: 0
  frame_id: /base_link
pose: 
  pose: 
    position: 
      x: -5.48213577271
      y: -3.74665260315
      z: 0.0
    orientation: 
      x: 0.0
      y: 0.0
      z: 0.457581479883
      w: 0.889167694683
  covariance: [0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.06853891945200942, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]