Multi-robot mission planning using evolutionary computation with incremental task addition

Abstract

Mission planning asks the robot to solve complex missions, requiring the robot to execute several actions such that a complex goal condition (called mission) is satisfied. The missions can be specified by using Linear Temporal Logic or any custom domain-specific language. Here, we consider service robots in the home or office environments with multiple users, each specifying a task that together makes a mission. Conventional mission planning techniques assume that the complete mission specification is known in advance, while mission planning is incremental wherein tasks keep getting added and deleted. The exponential computational complexity of the current model verification-based approaches is not a viable solution. The paper presents an evolutionary framework for incrementally solving the problem of mission planning for the problems for which a polynomial-time verification system exists for the tasks. The optimal solution is represented in an encoding-specific format that is compiled to form a linear trajectory of the robots. Optimization is done as the robot operates. Thus, the robot at any time has a partial solution to every task that has already been executed that (along with the robot assigned to the task) cannot be altered, while a part representing the next steps of the robot to be optimized. This acts as a constraint in the continuous optimization. The proposed approach uses Dynamic Programming to integrate the solutions of tasks and as a distance function in the diversity preserving evolutionary computation since the mission constantly changes. Comparisons are made with a baseline evolutionary computation without Dynamic Programming, a non-incremental version of the proposed algorithm, and several greedy strategies. The proposed algorithm is seen to perform better than all other methods. Further, the algorithm is implemented and demonstrated on a Pioneer LX robot.