Planejamento probabilístico sensível a risco com ILAO* e função utilidade exponencial

Abstract

Planejamento probabilístico sensível a risco com ILAO* e função utilidade exponencial São Paulo 2019Autorizo a reprodução e divulgação total ou parcial deste trabalho, por qualquer meio convencional ou eletrônico, para fins de estudo e pesquisa, desde que citada a fonte.

Heuristics based on projection occupation measures for probabilistic planning with dead-ends and risk

“…To the best of our knowledge, this is the first heuristic to risk-sensitive ssps that takes into account the probabilities and the risk factor. The empirical results show that a heuristic search algorithm with the proposed heuristic h rs pom outperforms the solution of Freitas (2019), considered the state-of-the-art for rs-ssps with exponential utility function.…”

“…In Appendix E we discuss different algorithms to find this extreme risk λ factor 1 . [Freitas, 2019] proposed the first heuristic search solution for risk-sensitive ssps based on expected exponential utility [Patek, 2001]. In that work, the algorithm ilao [Hansen and Zilberstein, 2001], a different implementation of the F&R framework (Algorithm 2),was implemented with a heuristic that considers the minimum local utility.…”

“…This heuristic, which we will call h rs local , considers the risk factor and the lowest immediate cost for each non-goal state. Thus, the heuristic h rs local is defined as follows [Freitas, 2019]: s,a) otherwise.…”

“…In Chapter 2 we have shown another asynchronous algorithm that is an implementation of the F&R framework, the lrtdp. Algorithm 6 modifies the algorithm lrtdp (Algorithm 3) for risk-sensitive ssp problems, updating the value function of the states visited in each trial according with Equation 5.3, assuming a λ-feasible value as input and a heuristic that can be the h rs local heuristic proposed by [Freitas, 2019], or any other heuristic that takes the risk into account (as the one we propose in Chapter 3). The remaining functions of the original lrtdp are the same.…”

“…• rs-lrtdp algorithm (lrtdp with Bellman update based on Equation 5.3) with heuristic h rs local [Freitas, 2019], that computes Equation 5.5;…”