How to improve AMCL pose estimate?

edit

Hi all,

While developing my own path planner, I found out that AMCL cannot yield acceptable pose estimate despite modifying all possible parameters. Here (with more detailed observation) I recorded a video to show how erroneous the estimation is. The platform used is a differential non-holonomic vehicle, the sensors are Hokuyo laser range finder with Asus Xtion camera at the front. And the AMCL configuration is as below assumed the argument tags' values are all correct (why can't it show the /node tag at the end?):

<node pkg="amcl" type="amcl" name="amcl">
<param name="use_map_topic"             value="$(arg use_map_topic)"/>
<!-- Publish scans from best pose at a max of 10 Hz -->
<param name="odom_model_type"           value="diff"/>
<param name="odom_alpha5"               value="0.1"/>
<param name="gui_publish_rate"          value="10.0"/>
<param name="laser_max_beams"             value="60"/>
<param name="laser_max_range"           value="12.0"/>
<param name="min_particles"             value="500"/>
<param name="max_particles"             value="2000"/>
<param name="kld_err"                   value="0.05"/>
<param name="kld_z"                     value="0.99"/>
<param name="odom_alpha1"               value="0.2"/>
<param name="odom_alpha2"               value="0.2"/>
<!-- translation std dev, m -->
<param name="odom_alpha3"               value="0.2"/>
<param name="odom_alpha4"               value="0.2"/>
<param name="laser_z_hit"               value="0.5"/>
<param name="laser_z_short"             value="0.05"/>
<param name="laser_z_max"               value="0.05"/>
<param name="laser_z_rand"              value="0.5"/>
<param name="laser_sigma_hit"           value="0.2"/>
<param name="laser_lambda_short"        value="0.1"/>
<param name="laser_model_type"          value="likelihood_field"/>
<!-- <param name="laser_model_type" value="beam"/> -->
<param name="laser_likelihood_max_dist" value="2.0"/>
<param name="update_min_d"              value="0.25"/>
<param name="update_min_a"              value="0.2"/>
<param name="odom_frame_id"             value="$(arg odom_frame_id)"/> 
<param name="base_frame_id"             value="$(arg base_frame_id)"/> 
<param name="global_frame_id"           value="$(arg global_frame_id)"/>
<param name="resample_interval"         value="1"/>
<!-- Increase tolerance because the computer can get quite busy -->
<param name="transform_tolerance"       value="1.0"/>
<param name="recovery_alpha_slow"       value="0.0"/>
<param name="recovery_alpha_fast"       value="0.0"/>
<param name="initial_pose_x"            value="$(arg initial_pose_x)"/>
<param name="initial_pose_y"            value="$(arg initial_pose_y)"/>
<param name="initial_pose_a"            value="$(arg initial_pose_a)"/>
<remap from="scan"                      to="$(arg scan_topic)"/>

</node>

I have tried other packages such as humanoid_localization which keeps complaining of receiving errorneous occupancy grid maps, or fused /amcl_pose with wheel odometry with robot_localization. I also looked into hector_pose_estimation but concluded that there are no matching topics that I can give to the package. Another option is to replace AMCL with MRPT_localization. MRPT localization node does publish odom and I will look into how to generate map of its format now. Despite this, the rest of my attempts end in failure.
Thus, how do you resolve this kind of localization problem? Or is there an easy or effective way to get around it? Is there any other alternative apart from using AMCL? Thank you !