#### DMCA

## Dynamic multi-agent task allocation with spatial and temporal constraints (2014)

Venue: | In AAAI |

Citations: | 1 - 0 self |

### Citations

126 | Market equilibrium via a primal-dual algorithm for a convex program
- Devanur, Papadimitriou, et al.
(Show Context)
Citation Context ...balanced over the agents. We demonstrate in Section 4 that this results in higher team utility in the full LEP than directly trying to maximize the utility represented by R. The FMC 1495 0 10000 20000 30000 40000 50000 0 100 200 300 400 500 Ut ili ty Time (minutes) FMC_TA SA LP Greedy CFLA+ SA+ Figure 1: Accumulated team utility for shifts with 60 events. 0 5 10 15 20 25 30 0 20 40 60 80 100A ve ra ge T im e (m in ut es ) # of Events FMC_TA SA LP Greedy CFLA+ SA+ Figure 2: Response times compared to centralized approaches. allocation can be computed in polynomial time in a centralized setting [2] or in pseudo-polynomial time in a distributed setting [6]. The final stage of FMC_TA schedules the fractions of tasks allocated to agents using a greedy heuristic that maximizes the utility that agents will derive from fulfilling a task. Each ai schedules its tasks in decreasing order of xijrij . This can be done independently by each agent in both centralized and distributed settings. 4. EXPERIMENTAL EVALUATION We compare FMC_TA to five centralized benchmark algorithms. Two are versions of simulated annealing: SA used a random starting point on every reallocation while SA+ started from the p... |

82 |
Distributed stochastic search and distributed breakout: properties, comparison and applications to constraint optimization problems in sensor networks
- Zhang, Wang, et al.
- 2005
(Show Context)
Citation Context ...vents. FMC_TA outperforms the other approaches in this respect as well, as seen in 0 5 10 15 20 25 30 0 20 40 60 80 100A ve ra ge T im e (m in ut es ) # of Events FMC_TA MGM-2 DSA DSAN-K Figure 3: Response times compared to distributed approaches. Figure 2. This is due to much faster responses to the more important type 1 and type 2 events, achieved by sharing more tasks than the other approaches (not shown for lack of space). FMC_TA is similarly effective when compared to distributed approaches. Figure 3 shows that FMC_TA achieves faster response times than three leading DCOP algorithms: DSA [7], Distributed Simulated Annealing (DSAN) [1], and MGM-2 [3]. Again, this is especially true of more important tasks, resulting in FMC_TA achieving higher team utility (omitted for lack of space). 5. CONCLUSIONS In this paper we proposed a new approach for dynamic task allocation that uses a simplified problem model to generate fair (envyfree) and efficient (Pareto optimal) allocations. We hypothesized that this combination of properties results in high quality solutions for task allocation problems in which we want all agents to contribute efficiently in order to achieve the group goal. Our ex... |

42 | Distributed Algorithms for DCOP: A Graphical-Game-Based Approach.
- Maheswaran, Pearce, et al.
- 2004
(Show Context)
Citation Context ...ect as well, as seen in 0 5 10 15 20 25 30 0 20 40 60 80 100A ve ra ge T im e (m in ut es ) # of Events FMC_TA MGM-2 DSA DSAN-K Figure 3: Response times compared to distributed approaches. Figure 2. This is due to much faster responses to the more important type 1 and type 2 events, achieved by sharing more tasks than the other approaches (not shown for lack of space). FMC_TA is similarly effective when compared to distributed approaches. Figure 3 shows that FMC_TA achieves faster response times than three leading DCOP algorithms: DSA [7], Distributed Simulated Annealing (DSAN) [1], and MGM-2 [3]. Again, this is especially true of more important tasks, resulting in FMC_TA achieving higher team utility (omitted for lack of space). 5. CONCLUSIONS In this paper we proposed a new approach for dynamic task allocation that uses a simplified problem model to generate fair (envyfree) and efficient (Pareto optimal) allocations. We hypothesized that this combination of properties results in high quality solutions for task allocation problems in which we want all agents to contribute efficiently in order to achieve the group goal. Our experiments support this hypothesis, demonstrating the advant... |

21 |
Coalition formation with spatial and temporal constraints.
- Ramchurn, Polukarov, et al.
- 2010
(Show Context)
Citation Context ...setting [6]. The final stage of FMC_TA schedules the fractions of tasks allocated to agents using a greedy heuristic that maximizes the utility that agents will derive from fulfilling a task. Each ai schedules its tasks in decreasing order of xijrij . This can be done independently by each agent in both centralized and distributed settings. 4. EXPERIMENTAL EVALUATION We compare FMC_TA to five centralized benchmark algorithms. Two are versions of simulated annealing: SA used a random starting point on every reallocation while SA+ started from the previous allocation. CFLA+ is a version of CFLA [4] adapted to LEP by computing the maximum utility for pairs of tasks taking into account the soft deadlines. Greedy allocates tasks in decreasing order of importance to the agent that would derive highest utility. LP is identical to FMC_TA except that it finds X by using a linear program that directly maximizes team utility as represented by R. We considered 20 random problems of an 8-hour shift with 9 agents in a city of 6 × 6 km divided into 9 neighborhoods. There were four types of events of decreasing importance from type 1 to type 4. Distribution of event types and workloads were based on ... |

7 | Proportional response dynamics in the fisher market,
- Zhang
- 2011
(Show Context)
Citation Context ... this results in higher team utility in the full LEP than directly trying to maximize the utility represented by R. The FMC 1495 0 10000 20000 30000 40000 50000 0 100 200 300 400 500 Ut ili ty Time (minutes) FMC_TA SA LP Greedy CFLA+ SA+ Figure 1: Accumulated team utility for shifts with 60 events. 0 5 10 15 20 25 30 0 20 40 60 80 100A ve ra ge T im e (m in ut es ) # of Events FMC_TA SA LP Greedy CFLA+ SA+ Figure 2: Response times compared to centralized approaches. allocation can be computed in polynomial time in a centralized setting [2] or in pseudo-polynomial time in a distributed setting [6]. The final stage of FMC_TA schedules the fractions of tasks allocated to agents using a greedy heuristic that maximizes the utility that agents will derive from fulfilling a task. Each ai schedules its tasks in decreasing order of xijrij . This can be done independently by each agent in both centralized and distributed settings. 4. EXPERIMENTAL EVALUATION We compare FMC_TA to five centralized benchmark algorithms. Two are versions of simulated annealing: SA used a random starting point on every reallocation while SA+ started from the previous allocation. CFLA+ is a version of CFLA [4] adapted... |

4 |
On finding an envy-free Pareto-optimal division.
- Reijnierse, Potters
- 1998
(Show Context)
Citation Context ...if vj = CTi or if CTi is a patrol. The utility of working on vj decreases with the time t when task execution begins, according to the soft deadline function, δ(vj , t) = βγ(t−α(t)), where β ∈ (0, 1] and γ ≥ 0 are constants; γ = 0 for patrols as they have no deadline. Utility for a single agent ai is U(ai) = ( Mi∑ k=1 xiskI(vsk )δ(vsk , tk) ) − π(vs1 ,∆w) and the total team utility is the sum of utilities for every agent. 3. FMC-BASED TASK ALLOCATION We propose an innovative task allocation algorithm, FMC_TA, based on Fisher market clearing (FMC). FMC_TA first creates a Fisher market instance [5] with agents as buyers and tasks as goods; agents are given equal monetary endowments. The matrix R specifying buyer preferences for goods is created by considering agent utility in a simplified task allocation problem that ignores inter-task spatial and temporal constraints. Specifically, rij = xijI(vj)− ρ(ai, vj)− π(CTi,∆w). FMC_TA next solves the Fisher market to get an allocation X that is envy-free and Pareto optimal with respect to the simplified problem [5]. These properties ensure that we achieve an efficient allocation that is balanced over the agents. We demonstrate in Section 4 that... |

2 |
Distributed simulated annealing. Distributed Constraint Problem Solving and Reasoning
- Arshad, Silaghi
- 2004
(Show Context)
Citation Context ...es in this respect as well, as seen in 0 5 10 15 20 25 30 0 20 40 60 80 100A ve ra ge T im e (m in ut es ) # of Events FMC_TA MGM-2 DSA DSAN-K Figure 3: Response times compared to distributed approaches. Figure 2. This is due to much faster responses to the more important type 1 and type 2 events, achieved by sharing more tasks than the other approaches (not shown for lack of space). FMC_TA is similarly effective when compared to distributed approaches. Figure 3 shows that FMC_TA achieves faster response times than three leading DCOP algorithms: DSA [7], Distributed Simulated Annealing (DSAN) [1], and MGM-2 [3]. Again, this is especially true of more important tasks, resulting in FMC_TA achieving higher team utility (omitted for lack of space). 5. CONCLUSIONS In this paper we proposed a new approach for dynamic task allocation that uses a simplified problem model to generate fair (envyfree) and efficient (Pareto optimal) allocations. We hypothesized that this combination of properties results in high quality solutions for task allocation problems in which we want all agents to contribute efficiently in order to achieve the group goal. Our experiments support this hypothesis, demonstra... |