Cooperative task planning considering co-work of robots and cooperators for clients

Abstract

This dissertation establishes an efficient minimum-time task planning algorithm to serve clients with robots and cooperators. For each task to serve a client, we require a preparatory moving action by a robot, followed by a simultaneous cooperative action by the same robot and a cooperator. As the number of clients is increased, a considerable number of nodes can be generated in the tree graph (NP-hard), similar to the multiple traveling salesman problem (mTSP). This dissertation is composed of three parts: 1) Extended representation of Task and Actions (ERTA), 2) Offline cooperative task planning, and 3) Online cooperative task planning. An offline cooperative task planning is divided to two problems. At first, we establish an efficient Minimum-time Cooperative Task Planning (MCTP) adopting a branch-and-bound which finds the minimum-time task schedule for a fixed number of robots (and cooperators). We also establish Cost-effective Cooperative Task Planning (CCTP), which finds the number of robots and the task schedule with the best Cost Performance Ratio (CPR). An online cooperative task planning is divided to two problems. First, we establish Emergency Cooperative Task Planning (ECTP) dealing with high-priority emergency tasks requiring interruption of the existing task schedule. We also establish Remaining Cooperative Task planning (RCTP) adopting a branch-and-bound, which can re-plan the existing task schedule for remnant normal tasks after Emergency Task schedule.

First, we propose Extended Representation of Tasks and Actions (ERTA) to formulate the task planning for serving robots and cooperators. Several robots cannot work with one cooperator at the same time. Hence the task planner always consider cooperator-conflict to find minimum-time task schedule. Our ERTA deals with concurrently executing tasks without cooperator-conflict.

Next, an offline cooperative task planning is divided to MCTP algorithm and CCTP algorithm. Our MCTP algorithm adopts branch-and-bound to find the minimum-time task schedule with a fixed number of robots. However, the conventional branch-and-bound can consume intractable amounts of storage and computing time because our task planning problem is NP-hard. For an efficient branch-and-bound, we propose initial tree using a Good Task Plan Candidates (GTPCs), and complementary heuristics considering robots and cooperators. They can reduce the tree size by intractable amounts. Our CCTP algorithm finds the most cost-effective task schedule with the minimum-CPR, and the most cost-effective number of robots. However, it is inefficient to iterate the task planning for the full search-range for finding the cost-effective task schedule. We propose an efficient procedure which reduces the search range via a relational analysis between the number of robots and the CPR. We use backward search for the reduced search range, and use the previous planning result to reduce the number of nodes.

Last, an online cooperative task planning deals with high priority emergency tasks requiring interruption of the existing task schedule. Our ECTP algorithm can quickly find emergency task schedule within 1 ms since we remove unnecessary search ranges by using Parallel Emergency Tasks (PET), Serial Emergency Tasks (SET), and Three Serial Emergency Tasks (TSET) algorithms. We also proposed the efficient RCTP algorithm based on branch-and-bound algorithm, which can quickly determine a schedule composed of the next R remaining task every serve duration.