|
|
The first experiment presented here consists of global localization.
The global localization process is outlined here.
The robot was placed in the environment near node one (N1).
The robot accessed the GVG and drove to node one, and was then given the atlas
for the environment that it generated in a previous experiment. Shown below are the RGVG and
feature maps superimposed on a figure of the environment.
For the global localization experiment, the robot was given no indication of its whereabouts.
The robot stored its observed node location (Ñ0)in its matching path and traversed its arrival edge
to its second observed node location (Ñ1). The search forest was then pruned according to the test
criterion, and the eliminated nodes are marked in the figure below: "T" for degree mismatch, "M" for minima equidistance
mismatch, "D" for edge length mismatch, and "F" for edge map mismatch.
As shown, the node tests were sufficient to identify which edge the robot had traversed (1:1,2:0) but not the direction
of travel along that edge. This is because the endpoint nodes 1 and 2 are similar and the edge length tests do no capture
directionality. The edge feature map test does subsume direction of travel, as the comparisons shown below illustrate.
The match path edge map features are shown as circles, and the candidate edge map features as crosses.
The match and true edge traversed was node 1 edge 1, arriving at node 2 edge 0, and this is what the robot successfully reported.
 |
 |
 |
 |
The second experiment is a kidnapped robot scenario. The setup is mostly the same as in the prior experiment.
The difference is that when the robot arrives at node 1 on edge 1, it is given the atlas and false node
location data. Specifically, it is told that it is at node 1 having arrived on edge 2.
The robot is also given a goal node location of node 12.
The robot plans its path from node 1 to node 12, departing along what it believes to be edge 1:2. In truth,
the robot departs on edge 1:1 and arrives at node 2 on edge 0. Upon comparing node and edge data observed to
the expected node and edge data in the atlas, the robot concludes that it has not traversed edge 1:2 nor arrived
at node 12. Using the observed data as a matching path, the robot localizes itself as having driven to node 2
on edge 0, with the same search forest result as in experiment A above.
The robot then plans its path from node 2 to node 12, and departs on edge 2:2. Upon arriving at node 12, the node
and edge data observed match the expected atlas data, and the robot concludes that it has successfully arrived at
node 12 on edge 12:2.
Go back to: H-SLAM Experiments Page | H-SLAM page
Jump to: People | Robots | Research | Events | Projects | Education | Papers | Software | Links
© Copyright 2002 Sensor Based Planning Lab, Carnegie Mellon University. All Rights Reserved.
 |
 |
 |