System for combining plurality of input control policies to provide a compositional output control policy

An article of manufacture comprising a data storage medium tangibly embodying a program of machine-readable instructions executable by a digital processing apparatus to perform method steps for combining policy information, the method steps comprising:

• (a.) Identifying v sub-policies at time t:
• {πm,it}, i=1,2, . . . ,v, where πm,it is the policy function for the ith sub-policy at time t, for t=0, 1, 2, . . . ,
• (b.) Specifying a set of actions A where the number of actions in A is denoted by #A and actions in A are denoted by aεA, and specifying a stimulus stεS,
• (c.) Specifying a set of permissible mature distributions over v-dimensional policy space:
• E⊂Iv, where I is the real-valued interval from 0 to 1 inclusive (i.e., including 0 and 1 as endpoints), and Iv is the v-dimensional space obtained by taking cross-products of I,
• (d.) Specifying a v-dimensional “recursive mixing function” gm:
• S×E→E, such that for the stimulus stεS, and mixing value hεE, gm(st,h)εE, where each dimension of said gm() is denoted by gmi(), i=1,2,. . . ,v,
• (e.) Specifying a value of the recursive mixing function at the previous time step t−1 represented by the recursive function ht=gm(st−1,ht−1), such that the recursion is finite such that for t=0, ht=h0 and h0 is defined to take a value in E,
• (f.) Computing a functional composition of the v sub-policies {πm,it}, i=1,2, . . . ,v, and the v-dimensional recursive mixing function gm(st, ht), given the stimulus st and the previous mixing value ht,
• whereby the plurality of v sub-policies can be subsequently combined according to said mixture distributions.

A method of combining a plurality of input value functions, comprising:

• (a.) providing an input information transmitting device representing an input control stimulus,
• (b.) providing an input information transmitting device representing a plurality of input value functions,
• (c.) providing an output information transmitting device representing an output value function,
• (d.) combining said input value functions into said output value function, such that more than one said input value function may simultaneously influence said output value function for said input control stimulus,
• (e.) transmitting said output value function via said output information transmitting device,
• whereby said method will combine said input value function by functional composition to obtain a single output value function,
• whereby more than one said input value function are able to simultaneously influence the said output value function for the said input control stimulus,
• whereby said method can iterate over time,
• whereby said method will allow said output value function to smoothly transition control from being influenced substantially by one of said input value functions to being influenced by substantially another of said input value functions,
• whereby said method will allow said output value function to smoothly transition control from being influenced substantially by one particular functional composition of said input value functions to being influenced substantially by another functional composition of said input value functions,
• whereby said method will allow the combination of a plurality of input value functions for the purpose of consolidating that information into a form suitable for use by a policy-based “action selection executive” (i.e., a policy-based “controller”) that is able to convert a value function into a control policy, and then use that control policy to automate the execution of action selection.