Dynamic programming accelerated evolutionary planning for constrained robotic missions

Abstract

Attributed to the increased automation, the day is not far wherein the robots will be seen doing a lot of sophisticated tasks, after which it is imperative that the offices and homes will have robots to replace the secretaries to be of common use for a large number of office-mates or house-mates. A mission comprises of a collection of high order tasks that a robot is asked to do with some logical and temporal constraints. The current approaches using model verification techniques have exponential complexity in terms of the number of variables, and are therefore not scalable to a very large level. The paper proposes a constrained mission specification language consisting of a sub-task as a logical relation between atomic tasks, a task as a collection of tasks to be performed one after the other, and a mission consisting of multiple tasks given by different users. An evolutionary approach is used to compute the solution to the mission that can scale to a very large number of variables. Problem specific heuristics are devised to compute a solution quickly. Particularly Dynamic Programming is used to align the solutions of multiple tasks to make a solution of a mission. Experimental results confirm that the proposed solution performs extremely well as compared to exhaustive search based approaches, model verification approaches and evolutionary approaches available in the literature. The results are demonstrated in simulations and on the Pioneer LX robot in the lab arena.