real time neural network learning

[leeguy92]

hi, ive done experiments with neural networks before involving simulation. I have been looking for a way for a robot to learn whilst it is acting and improve itself, possibly adapting to slight environmental changes.

I think this is called real time learning.

are there any ways this can be done, aside from just letting the robot fail loads of times like you would in a simulator with a genetic algorithm or particle swarm optimisation, because that could take a while?

[Webbot]

You may be able to run a simulation of the network on your PC. Leave it running for hours. And then upload the network settings (weightings) etc to the robot so that it will have learned what the PC has learned.