Collective Decision-Theoretic Planning for Robot Platoon Formation

The robot platooning problem has been studied extensively by the robotics community under some assumptions such as communication existance and global full observability. In this paper, we consider the platooning problem where the previous assumptions are not valid. In such a context, platooning can be considered as a specific flocking which is a collective decision model. This model can, thus, be seen as a decentralized multi-criteria decision making process. Vector-Valued Decentralized Markov Decision Process (2V-DEC-MDP) is an interesting framework for multi-criteria collective decision. It has been shown that 2V-DEC-MDP does not consider communication, local interactions and use local full observability which is a sub-class of partial observability. In this paper, we adapt this framework to consider the notion of leader and the relationship with the stochastic games. The theoretic concept of optimality used in such contexts is the Stackelberg Equilibrium (SE). We give the assumptions under which the leader follows the SE when using 2V-DEC-MDP. We present, then, the adaptation of the initial value functions of the 2V-DEC-MDP, in order to reach an SE. Experiments shown us that using the initial 2V-DEC-MDP leads to a near SE with a weak complexity while the adapted 2V-DEC-MDP leads to a SE with a very high complexity and thus a limited scalability which limits its applicability in real-life robotic applications. Mots-clés : Multiagent Planning, Markov Decisions Processes, Game Theory, Multi-Robot Systems