Heterogeneous Decision-Theoretic Autonomous Fleet of Operators and Robots for Recognition, Surveillance and security of goods and persons – GARDES

The aim of this project is to explore new research directions in terms of architectures, models and algorithms using decision-theoretic and game-theoretic approaches to control operational unit. This unit is composed on autonomous robots navigating in coalition and in coordination with controled robots (tele-operated) with unknown strategies of navigation in support of a person group (slodiers) to patrol and secure their immediat neighborhood. For such systems, GARDES project aims at develping a hierarchical interaction architecture. At the high level of the interaction: autonomous robots and slodiers can interact allowing soldiers to use robots (mode semi-autonomous) by sending them advice. At the low level, autonomous robots and semi-autonomous (controled, tele-operated robots) interact to coordinate their navigation.

GARDES aims at exploring decision-theoretic approaches allowing robots to be autonomous and accomplishing a mission and to interact with soldirs for assistance. To this end, the scientific challenges of GARDES are :
1. Multi-robot adjustable autonomy : How autonomous robots take into account the advice of the soldiers. This objective will be based on an approach called Mixed Markov Decision Process (MI-MDP) which allows a robot to decide not only on what to do but when to consider the advice of the soldiers. In GARDES project, we will consider an extension of this model to multi-robot settings and developing a Mixed Decentralized MDPs (MI-DEC-MDP). This is an innovative point of the project.
2. Heterogeneous interaction between autonomous and semi-autonomous robots: This objective concerns the problem of how an autonomous robot computes a coordinated strategy (policy) with semi-autonomous robots with lack of information on their strategies. The DGA PhD thesis of Arnaud Canu is a starting point where a DEC-POMDP has been proposed and we aim at extending this model for limited observability and communication constraints. Also, we consider adapt this approach to a stochastic game by considering semi-autonomous robots as leaders with unknown strategies and autonomous robots as followers which should compute their strategies accordingly. This is an innovative point of the project.
3. Control of an operational unit : This objective consists of considering adjustable autonomy in stochastic games where an autonomous robot should consider the advice of the soldiers to compute mixed policies that should be coordinated by estimated policies of semi-autonomous robots. To our knowledge, considering adjustable autonomy from game theory point of view is novel .

This project is not only about fundamental research issues, but it targets scnearios using real robots and considering operational staff. In GARDES, it is envisionned to develop a system and to integrate it in real platforms like µtrooper and Nerva robots developed by Nexter.