Self-organizing maps for path planning of a SCARA robot
This thesis studies the applicability of the Self-Organizing Maps (SOM) in generating the necessary joint space coordinates for robotics path planning.A survey of literature about robotics path planning and neural networks, more specifically, Self-Organizing Maps was conducted. The problem of how to...
Saved in:
Main Author: | |
---|---|
Format: | text |
Language: | English |
Published: |
Animo Repository
2000
|
Subjects: | |
Online Access: | https://animorepository.dlsu.edu.ph/etd_masteral/2346 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | De La Salle University |
Language: | English |
Summary: | This thesis studies the applicability of the Self-Organizing Maps (SOM) in generating the necessary joint space coordinates for robotics path planning.A survey of literature about robotics path planning and neural networks, more specifically, Self-Organizing Maps was conducted. The problem of how to approximate a Cartesian path to be followed by the end-effector of a robot was investigated. Possible methodologies for path approximation based on the SOM and linear interpolation was formulated. Thereafter, the SOM training procedures and the approximation methods were implemented. SOM of various sizes were trained to determine the appropriate values for the training parameters and to ensure that proper map organization is accomplished.The approximation methods were tested via simulation. The simulation was conducted considering the dimensions of an actual SCARA robot. Approximation of primitive paths, specifically, straight-line paths and circular arcs were carried out. The approximation results were then analyzed based on the Euclidean distance between the theoretical and approximated points as the basis.In the study, a SOM based path planner for a SCARA robot was designed and implemented to use three approximation methods, namely approximation using nearest node, rectangular patches and triangular patches.
Because these approximation methods require proper SOM organization, the proper training parameters, a and y, were determined. The parameter a represents the percentage the map values are brought closer to the training data and y represents the neighborhood parameter, which correlates to the training radius. The unsupervised learning scheme was used because its performance was comparable to that of the supervised learning scheme. Thus, the values of the joint coordinates, 01 and 02 are the only information which had to be stored in the SOM nodes.Simulation was done for approximation of straight lines and circular arcs, the primitive paths of a robot to determine if the approach was applicable. The Euclidean distance between the ideal sampling point and the approximated point was used as a criterion of the accuracy of the approximation method.The simulation results revealed that the approximation using nearest node is not a feasible method because of the large discrepancy between the theoretical and approximated path. Approximation using rectangular and triangular patches prove to be applicable for use with SOM of reasonable size. In both cases, the errors were deemed acceptable for most real life industrial applications. It was also noted that the equations for the triangular patch approach are linear while those for the rectangular approach are quadratic. Thus, the triangular patch approach is simpler, faster, and more suitable for real-time applications.In conclusion, the SOM is applicable for use in the 2D path planning of a SCARA robot, although further research is needed to refine the methodology. |
---|