Neural Network Controller for a Mobile Robot

This video shows a novel unsupervised learning algorithm building and training an artificial neural network which is controller a mobile robot simulator.

Category:

Science & Technology

Tags:

• neural
• network
• mobile
• robot
• controller

[PsionNinja]

Sweet demo. One question, how are the neurons connected to each other code-wise?

I can program a neuron/node to "fire" and slowly decay and not fire again until the node is below the potential threshold.

The trick for me is to find a way to connect 2 or more nodes to cause a cascade without having to hard-code everything cause I want to create expanding nodes like your program does.

[TheCCI]

The neurons of the video use C++ class pointer lists to store the neural network

structure (nodes and connections). The pointers refer to presynaptic and

postsynaptic neurons (actually there is a connection object between). So, a

neuron has direct access to the data of other relevant neurons. New neurons

create new objects, which are again linked together with pointers. You may

want to create inhibitory connections for competitive neurons. Otherwise you

may not get useful spiking patterns.

[vivalatip]

Looks really impressive, I have never seen a growing network this way. Have worked with NEAT and HyperNEAT before, but this looks a lot better. Do you have an article or some more information about what it is doing? I understand when there is a new situation you add a new neuron. But how do you change weights and what are the rules for adding new neurons and what about connections? How does it become intelligent? Please contact me, my e-mail: pietergbosma [at] hotmail [dot] com

Nice demo!

[essence324]

it should have more memory to expanse the network and capabilities.