Learning-enhanced market-based task allocation for oversubscribed domains

Abstract

Abstract-This paper presents a learning-enhanced marketbased task allocation approach for oversubscribed domains. In oversubscribed domains all tasks cannot be completed within the required deadlines due to a lack of resources. We focus specifically on domains where tasks can be generated throughout the mission, tasks can have different levels of importance and urgency, and penalties are assessed for failed commitments. Therefore, agents must reason about potential future events before making task commitments. Within these constraints, existing market-based approaches to task allocation can handle task importance and urgency, but do a poor job of anticipating future tasks, and are hence assessed a high number of penalties. In this work, we enhance a baseline market-based task allocation approach using regression-based learning to reduce overall incurred penalties. We illustrate the effectiveness of our approach in a simulated disaster response scenario by comparing performance with a baseline market-approach.

Biologically inspired redistribution of a swarm of robots among multiple sites

“…In the multi-robot domain, existing methods for the solution of these combinatorial optimization problems often reduce to market-based approaches (Lin and Zheng 2005;Guerrero and Oliver 2003;Jones et al 2006) where robots must execute complex bidding schemes to determine the appropriate allocation based on the various perceived costs and utilities. While market-based approaches have gained much success in various multi-robot applications Vail and Veloso 2003;Gerkey and Mataric 2002;Jones et al 2007) and can be further improved when learning is incorporated (Dahl et al 2006), these methods often scale poorly in terms of team size and number of tasks (Dias 2004;Golfarelli and Maio 1997).…”

Flood Disaster Mitigation: A Real-World Challenge Problem for Multi-agent Unmanned Surface Vehicles

“…As mentioned above, there exists many potential approaches to address our task allocation problem, that range from approximate DCOP solution techniques [21,5,11,4], to decomposing the problem as mixed integer linear programming problems [12,15], market based approaches [9,16], hybrid approaches [14,13], etc and that have been used in similar application domains. Despite the fairly rich suite of algorithms for addressing team planning, the dynamic and complex environments, continuous configuration and observation spaces, and relative large team sizes coupled with limited computing and sensing far exceed the complexity handled by many existing approaches.…”

Imitation learning for task allocation

“…In an oversubscribed disaster-response scenario, Jones et al enhance task allocation by modifying agent bids to better predict task penalties assessed for failing to complete tasks [13]. In their approach, learning was used to model the underlying causes for disparities between expected and actual rewards the agents receive, something which would be hard to explicitly model a priori.…”

“…In order to steer the task allocation mechanism towards the demonstrated allocations, we introduce a bias term into bids, so that the profit used for bidding is now [13]:…”