Title:
A control policy deciding device, a control policy deciding method, a control policy determination program, and a control system
Document Type and Number:
Japanese Patent JP6114679
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To provide a control measure determination device capable of performing high-speed approximation of a value iteration calculation.SOLUTION: A control measure determination device comprises: an initial linear function generation section 1 which generates a candidate group of linear functions which give linear components to a value function in a belief space on the basis of environment sensing information including indeterminacy; a dual transformation section 2 which transforms the candidate group in the belief space into a plurality of points in a dual space; a convex hull approximate calculation section 32 which calculates an approximate convex hull of the plurality of points; a membership determination section 34 which determines a membership function of apexes of the approximate convex hull; a convex hull upper side extraction section 36 which extracts an upper side of the approximate convex hull; an inverse dual transformation section 4 which inversely transforms the apexes belonging to the upper side into the linear function in the belief space; a linear function updating section 6 which updates the linear function in accordance with back-up step numbers on the basis of the obtained linear function and outputs the updated linear function to the dual transformation section 2; and a value function determination section 5 which obtains a plurality of linear components of an approximate value function on the basis of the linear function obtained through the inverse transformation after updating the linear function.
More Like This:
Inventors:
Hiroshi Tsukahara
Abe Mitsuru
Masato Obayashi
Abe Mitsuru
Masato Obayashi
Application Number:
JP2013235415A
Publication Date:
April 12, 2017
Filing Date:
November 13, 2013
Export Citation:
Assignee:
Denso IT Laboratory Inc.
International Classes:
G06N99/00; G06N7/00
Other References:
Hao Zhang,"Partially Observable Markov Decision Processes: A Geometric Technique and Analysis",Operations Research,2010年,第58巻,第1号,pp.214-228
南 泰浩,「部分観測マルコフ決定過程に基づく対話制御」,日本音響学会誌,社団法人日本音響学会,2011年,第67巻,第10号,pp.482-487
南 泰浩,「部分観測マルコフ決定過程に基づく対話制御」,日本音響学会誌,社団法人日本音響学会,2011年,第67巻,第10号,pp.482-487
Attorney, Agent or Firm:
Mamoru Suzuki
Shinji Kato
Shinji Kato