Least Action Sequence Determination in the Planning of Non-Prehensile Manipulation with Multiple Mobile Robots Changxiang Fan, Shouhei Shirafuji, and Jun Ota The University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa City, Chiba Prefecture, Japan fan@race.u-tokyo.ac.jp Abstract. To complete a non-prehensile manipulation task, using the lowest number of manipulation sequences with a well-determined num- ber of robots is desirable to improve efficiency of the manipulation and to bring stability to it. Since many possible states exist in the manipulation, various manipulation sequences exist to finish the task. Furthermore, the number of robots should be determined according to the environment. In this work, a graph-based planning method was used to determine the lowest possible number of sequences in the non-prehensile manipulation of mobile robots with a determined number of robots for gravity closure. Based on all possible object-environment contacts, the required number of robots for gravity closure was determined to generate possible ma- nipulation states in the contact configuration space. A state transition graph was created by representing the obtained states as nodes and then determining the least action sequences by searching for the shortest path in the graph. Keywords: Multiple mobile robots, non-prehensile manipulation, ma- nipulation planning, state transition graph 1 Introduction In robotic manipulations, it is often seen that objects are manipulated by ma- nipulators using prehensile method. However, such manipulation method is un- available for big-sized objects in narrow space, as big industrial robots can- not enter such environment. Therefore, the small-sized mobile robots are widely adopted to such kind of manipulation tasks owing to their motion flexibility. As the small mobile robots cannot manipulate big objects with prehensile method, non-prehensile method is adopted, in which an object is operated by pushing, pulling, or other similar actions [1]. Manipulations via multiple-robot systems must be effectively designed to perform the task cooperatively and efficiently. As non-prehensile manipulation is often conducted by using the environment that the targeted object lies in, the interaction between the targeted object and other objects in the surroundings should be considered in the planning of the manipu- lation. Maeda and Arai [2] discretized the contact configuration space of objects IAS-15 Proceedings of the 15th International Conference on Intelligent Autonomous Systems IAS 15 Baden-Baden, Germany, June 11th-15th 2018