Hi, I guess you are from robotic field. There, your robot has to do a certain task, e.g. "walk". To achieve an optimal walk you define some criteria (define the optimisation problem). Then you apply machine learning (your personal flavour, genetic programming artificial neural networks or other stuff) to find a good function to optimize your problem. A robot is presumably using sensors s and actuators a - you learn the relation a=C(s) as a closed loop to control your robot and perform the walk task.
Hi, It is an interesting question and it so happens that me and colleagues are investigating the inclusion of Neural Networks in the control algorithm design.
The neural network linearizes and estimates the system dynamics and helps in cancelling out the non-linearities of the system. The weight update equation though has to be managed in order to ensure the stability of the entire system.
Below are some of the papers that you can read for your reference:
1. I. Ranatunga, S. Cremer, F. L. Lewis, and D. O. Popa, “Neuroadaptive control for safe robots in human environments: A case study,” in Automation Science and Engineering (CASE), 2015 IEEE International Conference on, pp. 322–327, IEEE, 2015.
2. F. L. Lewis, K. Liu, and A. Yesildirek, “Neural net robot controller with guaranteed tracking performance,” IEEE Transactions on Neural Networks, vol. 6, no. 3, pp. 703–715, 1995.
3. F. L. Lewis, “Neural network control of robot manipulators,” IEEE Expert, vol. 11, no. 3, pp. 64–75, 1996.
4. S. Cremer, I. Ranatunga, S. K. Das, I. B. Wijayasinghe, and D. O. Popa, “Neuroadaptive calibration of tactile sensors for robot skin,” in Automation Science and Engineering (CASE), 2016 IEEE International Conference on, p. 1079–1085, IEEE, 2016.