A POMDP Framework for Dynamic Task Allocation and Reconnaissance of Multiple Unmanned Aerial Vehicles
Dongho Kim, Jae-Song Lee, Jaedeug Choi, Kee-Eung Kim · Jeongbo gwahaghoe nonmunji. so'peuteuweeo mich eung'yong · 2012
Interest in unmanned aerial vehicles(UAVs) for military use such as surveillance and target tracking or civil applications such as agriculture and firefighting has increased in recent years, due to significant progress in their development. Since there are still required human operators to remote-control multiple UAVs, autonomous control of UAVs has been significantly studied. Autonomous planning and control of multiple UAVs can be viewed as a sequential decision problem, which can be modeled as Markov decision processes (MDPs) or partially observable Markov decision processes (POMDPs) which extend MDPs by allowing partial or uncertain observations. This paper discusses the application of POMDPs to optimize the control of UAVs in reconnaissance flights with dynamic task allocation, and we experimentally show that the policies computed from POMDPs perform better than those from MDPs under uncertain observations. The applicability of the POMDP framework is also demonstrated by the realistic simulation using multiple quadcopters.