Verification of Markov Decision Processes with Risk-Sensitive Measures

被引:0
作者
Cubuktepe, Murat [1 ]
Topcu, Ufuk [1 ]
机构
[1] Univ Texas Austin, Dept Aerosp Engn & Engn Mech, 201 E 24th St, Austin, TX 78712 USA
来源
2018 ANNUAL AMERICAN CONTROL CONFERENCE (ACC) | 2018年
关键词
PROSPECT-THEORY;
D O I
暂无
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
We develop a method for computing policies in Markov decision processes with risk-sensitive measures subject to temporal logic constraints. Specifically, we use a particular risk-sensitive measure from cumulative prospect theory, which has been previously adopted in psychology and economics. The nonlinear transformation of the probabilities and utility functions yields a nonlinear programming problem, which makes computation of optimal policies typically challenging. We show that this nonlinear weighting function can be accurately approximated by the difference of two convex functions. This observation enables efficient policy computation using convex-concave programming. We demonstrate the effectiveness of the approach on several scenarios.
引用
收藏
页码:2371 / 2377
页数:7
相关论文
共 35 条
[31]  
Ratliff L. J., 2017, ARXIV170309842
[32]   Risk-averse dynamic programming for Markov decision processes [J].
Ruszczynski, Andrzej .
MATHEMATICAL PROGRAMMING, 2010, 125 (02) :235-261
[33]  
Shen XY, 2016, IEEE DECIS CONTR P, P1009, DOI 10.1109/CDC.2016.7798400
[34]  
Tamar A., 2015, 29 AAAI C ART INT
[35]   ADVANCES IN PROSPECT-THEORY - CUMULATIVE REPRESENTATION OF UNCERTAINTY [J].
TVERSKY, A ;
KAHNEMAN, D .
JOURNAL OF RISK AND UNCERTAINTY, 1992, 5 (04) :297-323