1 / 7

Neural network learning of Robot Navigation Tasks

Neural network learning of Robot Navigation Tasks. Megan DiVall ECE 539 Dec. 14, 2010. The “what” of the project. Use neural network classification tools studied in the class on a real set of data Compare performance of tools to each other The data:

kostya
Télécharger la présentation

Neural network learning of Robot Navigation Tasks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Neural network learning of Robot Navigation Tasks Megan DiVall ECE 539 Dec. 14, 2010

  2. The “what” of the project • Use neural network classification tools studied in the class on a real set of data • Compare performance of tools to each other • The data: • UC-Irvine Machine Learning Repository – “Wall Following Robot Navigation Data Set” • Donated by researchers at Federal University of Ceará, Brazil

  3. The “Why” of the project • Opportunity to apply lessons from class to a “real-life” problem in field of interest • Compare performance of different tools using the same set of realistic data • Compare performance of tools from class to those used in associated study • Perceptron performed poorly without short-term memory mechanisms, problem is not linearly separable

  4. Experimental procedure • Research/choose tools to use • Format data to be usable by each tool; create training/testing groups • (If needed) Modify tool’s programming/settings to produce good results • Perform tests noting classification rate, ease of use, speed of calculation, etc.

  5. Chosen Tools • Perceptron • Just plain perceptron; won’t work well if problem is not linearly separable • Multilayer Perceptron • Used in original study • K-nearest neighbor classifier • Not used in original study • Maximum likelihood classifier using uni-variate Gaussian model • Not used in original study

  6. Expected Results • Perceptron will not do well; original study found problem to not be linearly separable • Other tools may or may not do well but probably better than the perceptron • Multilayer perceptron did well in original study • One or two tools will prove superior both in classification rate and calculation ease/speed

  7. Discussion • What tool would I be most likely to use if I was programming a real robot? • Would the performance of the “best” tool be good enough for real applications? • Could anything be done to improve performance of the “best” tool? • How do my results compare to expected real-world robot navigation performance?

More Related