Data Storage in the Cerebellar Model Articulation Controller (CMAC)

Abstract

The storage of manipulator control functions in the CM AC memory is accomplished by an iterative process which, if the control function is sufficiently smooth, will converge. There are several different techniques for loading the CM AC memory depending on the amount of data which has already been stored and the degree of accuracy which is desired. The CM AC system lends itself to a “natural” partitioning of the control problem into manageable subproblems. At each level the CM AC controller translates commands from the next higher level into sequences of instructions to the next lower level. Data storage, or training, is accomplished first at the lowest level and must be completed, or nearly so, at each level before it can be initiated at the next higher level.

System Identification Using Fuzzy Cerebellar Model Articulation Controllers

“…As mentioned in the previous section, the traditional CMAC model (Albus, 1975a(Albus, , 1975b has fast learning ability and good local generalization capability for approximating nonlinear functions. The basic idea of the CMAC model is to store learned data in overlapping regions in a way that the data can easily be recalled yet use less storage space.…”

“…Being an artificial neural network inspired by the cerebellum, the cerebellar model articulation controller (CMAC) was firstly developed in (Albus, 1975a(Albus, , 1975b. With the advantages such as fast learning speed, high convergence rate, good generalization capability, and easier hardware implementation (Lin & Lee, 2009;Peng & Lin, 2011), the CMAC has been successfully applied to many fields; for example, identification (Lee et al, 2004), image coding (Iiguni, 1996), ultrasonic motors (Leu et al, 2010), grey relational analysis (Chang et al, 2010), pattern recognition (Glanz et al, 1991), robot control (Harmon et al, 2005;Mese, 2003;Miller et al, 1990), signal processing (Kolcz & Allinson, 1994), and diagnosis (Hung & Wang, 2004;Wang & Jiang, 2004).…”

Light-Weight Reinforcement Learning With Function Approximation for Real-Life Control Tasks

“…The use of function approximation rather than binary lookup-tables signifies that more than one feature value φ can be different from zero. This is also the case with the binary CMAC (Cerebellar Model Articulation Controller) neural network (Albus, 1975) used in (Singh and Sutton, 1996). When using CMAC, there are as many active features as there are overlapping layers in the CMAC, which could lead to excessive weight changes in Equation 1.…”

Biped dynamic walking using reinforcement learning

“…The Cerebellar Model Arithmetic Computer (CMAC) [1,2] is an associative neural network that tries to mimic the biological sensory neurones in the brain's cerebellum.…”