《动态任务分配》由会员分享,可在线阅读,更多相关《动态任务分配(10页珍藏版)》请在金锄头文库上搜索。
1、International Journal of Advanced Robotic Systems, Vol. 6, No. 4 (2009) ISSN 1729-8806, pp. 309-318 309Dynamic Task Allocation in Cooperative Robot Teams Dynamic Task Allocation in Cooperative Robot Teams Athanasios Tsalatsanis, Ali Yalcin and Kimon. P. Valavanis Center for Evidence-based Medicine a
2、nd Health Outcomes Research, University of South Florida, Tampa, FL USA Department of Industrial and Management Systems Engineering, University of South Florida, Tampa, FL USA Department of Electrical and Computer Engineering, University of Denver, Denver, CO USA E-mail: atsalatshealth.usf.edu) Abst
3、ract: In this paper a dynamic task allocation and controller design methodology for cooperative robot teams is presented. Fuzzy logic based utility functions are derived to quantify each robots ability to perform a task. These utility functions are used to allocate tasks in real-time through a limit
4、ed lookahead control methodology partially based on the basic principles of discrete event supervisory control theory. The proposed controller design methodology accommodates flexibility in task assignments, robot coordination, and tolerance to robot failures and repairs. Implementation details of t
5、he proposed methodology are demonstrated through a warehouse patrolling case study. Keywords: Supervisory control, cooperative robot teams, task allocation, limited lookahead policy. 1. Introduction Many applications in industrial, civilian and military fields benefit from mobile robot utilization.
6、Application domains vary from warehouse patrolling to service robotics and to space exploration. Mobile robots have been reported to explore, map or inspect friendly or hostile territories (Kumar the robot having the most relevant sensors for a task is assigned the particular task. Utility has also
7、been used in robot team cooperation to estimate the cost of executing an action (Botelho & Alami, 1999) and for sensor-based metrics (Gerkey & Mataric, 2002). Auction based approaches as in (Lagoudakis et al., 2004), (Botelho & Alami, 1999) and (Gerkey & Mataric, 2002) achieve task allocation based
8、on the Artificial Intelligence concept of Contract Net Protocol (Davis & Smith, 1983). Each robot bids for an available task and the robot with the higher bid is assigned to that task. In the proposed control methodology, the dynamic task allocation problem is addressed using utility and fuzzy logic
9、. Utility function values are computed based on the ability of each robot to perform a task considering several factors. Limited lookahead policies for supervisory control have been first studied in (Chung et al., 1992) where a limited lookahead window is used to control the online behavior of the u
10、ncontrolled system model. The notion of pending traces is introduced to describe the legality of a trace in the lookahead window based on a conservative or an optimistic attitude. The notion of pending traces was later raised in (Kumar et al., 1998) by extending the uncontrolled system model behavio
11、r by arbitrary traces beyond the limited lookahead window. In (Chung et al., 1993), the authors present a methodology that recursively computes the future control actions based on previously computed control actions. Later, in (Chung et al., 1994) and in (Hadj-Alouane et al., 1994) the authors prese
12、nt an extension to the lookahead policies to cope with the computational complexity problem by making a control decision without exploring the whole lookahead window. Further enhancements in limited lookahead policies for supervisory control have been proposed. In (Heymann & Lin, 1994) a lookahead p
13、olicy is presented for systems with partial observability. Also, in (Kumar & Garg, 2001) systems uncertainty is considered by assigning probabilities to event occurrences and in (Lin, 1993) by modeling all possible variations of the system. To our knowledge there are no limited lookahead policies in
14、 the literature designed to control cooperative robot teams. As noted in (Grigorov & Rudie, 2006), only few approaches, as described in (Yi-Liang et al., 1997) and (Gordon & Kiriakidis, 2000), concentrate in time varying systems where system modules appear or disappear in time. In these approaches r
15、esource modules disappear only after the completion of assigned tasks. In this work, we relax this assumption by considering failures during task execution. In coordinated robot teams the concept of robot failures and repairs is important since a robot failure while executing a task will lead to an
16、incomplete mission unless the control model reassigns the task elsewhere. The lookahead policy presented in this paper considers robot failures and repairs to ensure mission completion. Supervisory control based approaches on discrete event system have been used by a number of researches to control mobile robot teams. However, although a limited amount of work considers robot failures, not much effort is found i