SBP Lab 2 Slam Experiment : 06.07.2003

Experiment of June 7, 2003 : SBP Lab 2

Conducted by Deryck Morales (deryck@alumni.cmu.edu)

This experiment was conducted in SBP lab 2 here at Carnegie Mellon

This is the second experiment to demonstrate results of a re-visited node comparison on the Scout2.
The environment was the same as that constructed for the May 24th experiment;
This experiment differs in that weak meetpoint elimination is turned off, resulting in a reduced GVG with 5 meetpoints.
The environment formed the shape of a lower case "b", and photos are shown below.

Environment Photo 0 Environment Photo 2 Environment Photo 3 The corresponding graph

Results:

The 'b' environment as sensed The resultant graph

Nodes 1 and 0 Mismatch

Nodes 2 and 0 Mismatch Nodes 2 and 1 Mismatch

Nodes 3 and 0 Mismatch Nodes 3 and 1 Match Nodes 3 and 2 Match

Nodes 4 and 0 Mismatch Nodes 4 and 1 Mismatch Nodes 4 and 2 Mismatch Nodes 4 and 3 Match

Nodes 5 and 0 Mismatch Nodes 5 and 1 Match Nodes 5 and 2 Mismatch Nodes 5 and 3 Match Nodes 5 and 4 Mismatch

Discussion

Note that I use the words "node" and "meetpoint" interchangeably; the meetpoint exists in the environment, and corresponds to a node in the graph computed by the software.

The comparison of nodes consists of the following tests:
1. The degree of the node
2. Average minima distance : this value captures a sense of environment scale for the meetpoint. (This experiment used an equivalence range of 6 inches)
3. Relative edge angles : the leave angles of the edges of a meetpoint relative to each other. (This experiment used an equivalence angle range of 9 degrees)

The robot accessed the GVG from the position shown in the environment photos above. The robot then homed in on the first meetpoint, probed and recorded the details as node 0.

The robot then proceeded out of the "tail of the b" and traversed the environment, creating the nodes in the order labelled.

Of note is node 2, where the robot sensed four minima and probed the four potential edge directions.
The probing refined the local map and the correct three minima were stored for node 2. You may notice this in the sensed map figure above.
Also note the dead reckoning error at the position of nodes 5 and 1. The robot's actual position was the same at each node.

Upon arriving back to node 1 on edge 1, the robot created node 5 and compared this newest node to the existing set.
The results of these comparisons are shown above. Note that the correct edge correspondence was calculated on Nodes 5 and 1.

Note that several matches were found, and several of these may be eliminated by analysis of node connectivities.
For example, matches of nodes 3 with 2 and 4 with 3 could be eliminated because the nodes are neighbors and by the encoder coordinates we may confirm that 3 != 2 and 4 != 3;
therefore these should not match.

Go back to: H-SLAM Experiments Page | H-SLAM page