A parallel rollout algorithm for wildfire suppression

Mauro Montenegro, Roberto López, Rolando Menchaca-Méndez, Emanuel Becerra, Ricardo Menchaca-Méndez

Resultado de la investigación: Capítulo del libro/informe/acta de congresoContribución a la conferenciarevisión exhaustiva

Resumen

In this paper, we formulate the problem of optimal Wildfire Suppression as an infinite horizon Decision Process (DP) problem where an agent (e.g., a robotic firefighter) decides which areas to intervene to extinguish the fire. The dynamics of the wildfire is modeled by a cellular automaton whose state at time k is defined as a bi-dimensional grid $$x:k$$ where each cell in this grid describes the state of a rectangular geographic region of the wildland. The proposed algorithm, which is based on a non-parametric reinforcement learning (RL) methodology, computes optimized control policies that determine the agent’s actions that minimize a cost function that aims to preserve most of the cells with trees. From a given state $$x:k$$, the proposed algorithms employs rollout to take advantage of heuristic solutions to approximate, in polynomial-time, the future cost function. Two different heuristics approaches were applied: A corrective-based model that only takes into account surrounding burning cells, and a predictive strategy that calculates a coefficient-based metric over nearby trees and empty cells. We implemented a parallel sampler using CUDA to simulate the trajectories that rollout generates. This parallel implementation allows us to increase the number of lookahead steps without incurring in large computing times. Our experimental results show that the rollout strategy outperforms the base heuristics and that effectively suppresses wildfires.

Idioma originalInglés
Título de la publicación alojadaTelematics and Computing - 9th International Congress, WITCOM 2020, Proceedings
EditoresMiguel Félix Mata-Rivera, Roberto Zagal-Flores, Cristian Barria-Huidobro
EditorialSpringer Science and Business Media Deutschland GmbH
Páginas244-255
Número de páginas12
ISBN (versión impresa)9783030625535
DOI
EstadoPublicada - 2020
Publicado de forma externa
Evento9th International Congress on Telematics and Computing, WITCOM 2020 - Puerto Vallarta, México
Duración: 2 nov 20206 nov 2020

Serie de la publicación

NombreCommunications in Computer and Information Science
Volumen1280
ISSN (versión impresa)1865-0929
ISSN (versión digital)1865-0937

Conferencia

Conferencia9th International Congress on Telematics and Computing, WITCOM 2020
País/TerritorioMéxico
CiudadPuerto Vallarta
Período2/11/206/11/20

Huella

Profundice en los temas de investigación de 'A parallel rollout algorithm for wildfire suppression'. En conjunto forman una huella única.

Citar esto