This work builds on a decentralized task allocation algorithm for networked agents communicating through an asynchronous channel. The algorithm extends the Asynchronous Consensus-Based Bundle Algorithm (ACBBA) to account for more real time implementation issues resulting from a decentralized planner. This work utilizes a new implementation that allows further insight into the consensus and message passing structures of ACBBA. This paper specifically talks to the comparisons between global and

more »

... cal convergence in asynchronous consensus algorithms. Also a feature called asychronous replan is introduced to ACBBA's functionality that enables efficient updates to large changes in local situational awareness. A real-time software implementation using multiple agents communicating through the user datagram protocol (UDP) validates the proposed algorithm. I. Introduction Teams of heterogeneous unmanned agents are regularly employed in complex missions including intelligence, surveillance and reconnaissance operations. 1-3 Of critical interest for successful mission operations is proper coordination of tasks amongst a fleet of robotic agents. Many different methods have been considered for enabling agents to distribute tasks amongst themselves from a known mission task list. Centralized planners, which rely on agents communicating their state to a central server, are useful since they place much of the heavy processing requirements safely on the ground, making robots smaller and cheaper to build. 4-10 Ideally, the communication links between all elements of the system (command station, autonomous vehicles, manned vehicles, etc.) are high bandwidth, low latency, low cost, and highly reliable. However, even the most modern communication infrastructures do not possess all of these characteristics. If the inter-agent communication mechanism has a more favorable combination of these characteristics compared to agent-tobase communication, then a distributed planning architecture offers performance and robustness advantages. In particular, response times to changes in situational awareness can be made significantly faster via distributed control than those achieved under a purely centralized planner. As a result, distributed planning methods which eliminate the need for a central server have been explored. 11-14 Many of these methods often assume perfect communication links with infinite bandwidth in order to ensure that agents have the same situational awareness before planning. In the presence of inconsistencies in situational awareness, these distributed tasking algorithms can be augmented with consensus algorithms 15-24 to converge on a consistent state before performing the task allocation. Although consensus algorithms guarantee convergence on information, they may take a significant amount of time and often require transmitting large amounts of data. 25 *