Self-Organizing, Smart and Safe heterogeneous robots fleet by collective emergence for a mission – SOS

The revolution of robotics and artificial intelligence now makes it possible to dream of ambitious applications.
In particular, the implementation of intelligently managed fleets of robots would provide innovative solutions to many concrete problems. Among these, the SOS project (Self-Organizing, Smart and safe heterogeneous robots fleet by collective emergence for a mission) focuses on forest fire detection using a fleet of aerial and ground robots.
It brings together three research teams from two laboratories (CRIStAL and CRAN) and one SME (Lynxdrone) with the aim of proposing, designing and developing a mechanism for intelligent management of heterogeneous robot fleets by collective emergence:
i) taking into account the specific characteristics of the robots
ii) adaptive to the dynamic and evolving environment
iii) as well as to the estimated and predicted health state of the robots (in terms of control actuators, localization sensors, battery charge, residual life, ...)
iv) and resilient to the occurrence of an incident,
in order to fulfill a defined mission.
We put ourselves under the following assumptions:
i) the system is composed of several mobile robots,
ii) the robots are heterogeneous: terrestrial or aerial, with varying load, energy, computation, perception, localization, control, decision, communication, ... capacities,
iii) environmental conditions could be difficult,
iv) the robots are subject to sensor and/or actuator failures,
v) the chosen architecture of the multi-robot aero-terrestrial system is a decentralized architecture: there is no master.
The research conducted will bring contributions, under a safe framework, on three axes in particular:
i) Intelligent and decentralized self-organization of fleets
This involves designing the individual behaviors of robots, whether air or ground, which have different capacities of perception and action depending on their equipment. These robots must determine by themselves the role they should take on in order to best contribute to the collective realization of the mission. For example, these robots must be able to adapt their communication vector according to the characteristics of the environment, or to become "information relays" instead of patrolling an area. The goal is to obtain a system capable of dynamically reorganizing itself to face unexpected events, while each one has only incomplete information.
ii) Robot control
The aim is to explore within the learning paradigm the development of reconfiguration schemes for robot control that take into account the health of the system as well as predictions of future failures, thus guaranteeing the completion of the mission with a certain level of performance in terms of stability and safety.
iii) Robot localization
On the one hand, we will efficiently and innovatively exploit the cooperation between robots to improve the accuracy, availability and safety of the estimation of the robots' position. UAVs can, for example, be used as GNSS receivers remote and redundant to those of AGVs. Indeed, while the accuracy or even the availability of the AGV GNSS measurements can be affected by poor reception of satellite signals due to nearby obstacles, the quality of the UAV GNSS measurements can be better due to their higher altitude, above the obstacles. Thanks to perception, AGVs can then position themselves relatively to UAVs. Conversely, their positions in situ, can allow the AGVs to have a better positioning relative to the environment (possibly a marker) and benefit the UAVs. On the other hand, it will be necessary to make the localization fault-tolerant by adding a diagnostic layer in charge of detecting and isolating the faulty measurements and/or sensors in order to exclude them from the multi-sensor fusion procedure. The combination of model-based and data-based (machine learning) approaches will be investigated.
The proposed solutions will be tested and evaluated in simulation and on real data.