The experimental results of using trained neural networks to regulate the pitch attitude of an underwater telerobot are presented. The neuromorphic control algorithm developed for teaching the networks is a direct application of the back-propagation entrainment method, with those modifications required to pose the problem in a control systems framework. The networks perform in the field as predicted by simulation results; however, it is observed that the complexity of the calculations required can create unacceptable delays when using a single serial microprocessor to compute the control. This problem becomes especially acute when scaling the architecture to more complex neural topologies. Special-purpose hardware, which directly implements the neural equations and, hence, realizes the benefits of the natural parallelism of these models, is seen as a necessary development for the effective use of neural controllers.<<ETX>>
[1]
Richard P. Lippmann,et al.
An introduction to computing with neural nets
,
1987
.
[2]
Bernard Widrow,et al.
Punish/Reward: Learning with a Critic in Adaptive Threshold Systems
,
1973,
IEEE Trans. Syst. Man Cybern..
[3]
Moshe Kam,et al.
Neuromorphic architectures for fast adaptive robot control
,
1988,
Proceedings. 1988 IEEE International Conference on Robotics and Automation.
[4]
Geoffrey E. Hinton,et al.
Learning internal representations by error propagation
,
1986
.
[5]
R. Lippmann,et al.
An introduction to computing with neural nets
,
1987,
IEEE ASSP Magazine.