Cerebellar Model Articulation Controller (CMAC) has some attractive features: fast
learning capability and the possibility of efficient digital hardware implementation.
These features makes it a good choice for different control applications, like the
one presented in this paper. Theproblem is to navigate a mobilerobot(e.g. a car)from
an initial state to a fixed goal state. The approach applied is backpropagation through
time (BPTT). Besides the attractive features of CMAC it has a serious drawback: its
memory complexity may be very large. To reduce memory requirement different variants
of CMACs were developed. In this paper several variants are used for solving the navigation
problem to see if using a network with reduced memory size can solve the problem efficiently.
Only those solutions are described in detail that solve the problem in an acceptable
level. All of these variants of the CMAC require higher-order basis functions, as
for BPTT continuous input-output mapping of the applied neural network is required.
