Simulated annealing for multi-robot hierarchical task allocation with flexible constraints and objective functions

In this paper we study a novel approach in networked robotics for optimal allocation with interchangeable objective functions, from minimizing the worst-case cost of any agent in a multi-robot team in time-critical missions, to minimizing the team usage of resources. We propose a general model for flexible mission planning, using hierarchical task networks as descriptive framework, the multiple traveling salesmen as optimization model, and distributed simulated annealing for solution search in very large solution spaces. This proposal does not discard viable solutions, hence the optimal one for the model may be eventually found. We briefly comment on the feasibility and usefulness of full replication of data in critical missions. We present a working implementation in simulation and preliminary plans for urban environment experiments, and also an implementation using adapted market-based techniques for comparison against our proposal. For our simulations we use a coverage problem, where our algorithm enables us to determine the best starting point for the robots, exploiting the flexibility of HTNs.