MLP NN for a redundant planar manipulator kinematics with an obstacle avoidance

Links to articles:

Neural Network for the identification of a functional dependence using data preselection. in: Neural Network World, 2021/2, pp. 109-124. ISSN 1210-0552, doi: 10.14311/NNW.2021.31.006. Open access.

MLP Neural Network for a Kinematic Control of a Redundant Planar Manipulator. in: Jaroslav Beran, Martin Bílek, Miroslav Václavík and Petr Žabka: Advances in Mechanism Design III. Proceedings of TMM 2020. Liberec, 2021.

Kinematics Control of a Redundant Planar Manipulator with a MLP Neural Network. in: Proc. of the International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME), 7-8 October 2021, Mauritius. This IEEE article describes more advanced solution of the same.

The explanation of this problem is on the video.

If you are going to solve this with a 3D redundant manipulator and you need a help with preparation of the data, possibly I can help you with the program. You can send me a mail: hlavac@fs.cvut.cz, or info@zpp.wz.cz.

In this .zip file, you can find sources for a demonstration program, used in the video. Requires the Lazarus IDE to be installed. Without any guarantee. If you prefer the translated program (for MS Windows), send me a mail.

But this is a one-purpose program. If you need to change parameters, or use it in a different task, you need to modify the source program.

Prepared functions:

• Set the parameters if necessary - the editing boxes allow changing the manipulator geometry. But program is tested with the default settings only. For bigger changes, you have to modify the source.
• Button "Generate data" - click and wait. Prepares data for training into memo. Use clipboard (Ctrl+A, Ctrl+C) to transfer data into Excel, then use "text to columns" and save it into csv file for the Matlab or other tools (Scilab?).
• Button "Show sample" - depending on the checkboxes, it shows a graphics example of a manipulator shape. Use "CLR" after few tests.
• Button "Prepare demo" - it will take a while. After the white dots in the shape of the (reduced) working area is shown, you can click to any of the white dots and it will show the shape of the arm to reach this position. The drawn sample is erased from the memory, if you click many times to the same point, it will draw nothing. Press "CLR" to redraw the white points to see if there are some data to draw. Note, that this is a different CLR.