Multi-Agent Path Planning Based on Task-Swap Negotiation

Symbolic path planning in autonomous robotic agents is the search, on a network of places and routes modeling the environment, for a sequence of routes which enables an agent to execute a set of interrelated tasks. When several agents are present in the environment, each having been assigned a set of tasks, the adoption of coordination mechanisms allows the task-execution costs to be decreased. The formulation of tasks includes resources, resource types, temporal and precedence constraints , production and use of objects. In this paper we propose a market-like negotiation protocol which extends the well-known Contract Net Protocol to domains where task swap is the only possible type of contract. After planning an initial path to execute its tasks, each agent uses this protocol to progressively decrease the path cost by swapping tasks with other agents. The agents we consider are heterogeneous and self-interested, and communicate with each other by broadcasting messages with a limited radius.