240 likes | 428 Views
Decentralized Task Allocation for Heterogeneous Teams with Cooperation Constraints. Presented by: Sin Ting, Angela, Lui ARC Centre of Excellence for Autonomous Systems. Han-Lim Choi , Andrew K. Whitten, and Jonathan P. How. Outline. Problem Paper Contribution
E N D
Decentralized Task Allocation for Heterogeneous Teams with Cooperation Constraints Presented by: Sin Ting, Angela, Lui ARC Centre of Excellence for Autonomous Systems Han-Lim Choi, Andrew K. Whitten, and Jonathan P. How
Outline • Problem • Paper Contribution • Consensus Based Bundle Algorithm • Task Decomposition / Scoring • Task Elimination • Results
Problem Allocation task of agents to tasks. Given a list of J tasks of K types, and I agents of M types. Find the mapping of tasks to a required set of agents that maximize some global reward. • Solo Task (Tracking) • Preferred Duo Task (Tracking) • Required Duo Task (Tracking and striking)
Problem Optimization:
Contribution • Extending Consensus Bundle Based Algorithm to number of agents on task. • Constraints due to requirements & preferences of cooperation. • Embedding cooperation constraints in the scorings structure. • Decentralized methods to eliminate invalid assignment.
Consensus Based Bundle Algorithm (CBBA) Consensus-Based Decentralized Auctions for Robust Task Allocation (2009, H.Choi, L.Brunet and J.P.How) Based off winning agents and winning scores. • Bundle • Consensus (Conflict Resolution) Each agent carries five vectors
CBBA 1) Bundle • are tasks added based on time. • list based on location in the assignment which incurs the largest score improvement.
CBBA 2) Conflict Resolution • the winning bid list. • the winning agent list. • d time stamps of the last information update. If is different between two agents, the agent that receives the message will update, leave, or reset the task assignment. If there are any changes, the bundle is released and the agent has to replan.
CBBA - Example Example & figures from MIT CBBA code
CBBA Example & figures from MIT CBBA code
Consensus Based Bundle Algorithm (CBBA) Advantages: • Polynomial time algorithm. • Scales well with network and numbers of tasks. • Accounts for inconsistent information among networked agents. Disadvantages: • Allocation only allows one agent per task
Task Decomposition & Scoring Task Decomposition for Preferred Duo Task
Task Decomposition & Scoring Task Decomposition for Required Duo Task
Task Decomposition & Scoring Constraint to not allow a single agent to bid for the same task twice. Constraint for required duo
Task Decomposition & Scoring Optimization:
Decentralized Task Elimination After CBBA converges, the subtask allocations for required duo tasks will either be: • Both subtasks are assigned. • Both subtasks are not assigned. • Only one subtask is assigned. Since is the same for all agents when CBBA converges, each agent can reconstruct the bundle of other agents and eliminate invalid tasks in their bundle.
Decentralized Task Elimination Final Proposed Algorithm:
Results Track & Eliminate Task. Task allocation improvement over baseline CBBA. Figure 1 from Paper
Results Figure 2&3 from Paper