Abstract
The use of mobile robots is growing in many fields of application. These robots can be used in harsh environments where access is dangerous or even impossible for humans. Unfortunately, we have to face many undesired events during these missions. These can significantly impact the mission in progress or the safety of the robot and its environment or the control loop of the robot.Dependability techniques provide proven solutions in many areas, and in particular fault tolerance which allows to detect and treat these situations at runtime. However, the proposed approaches lack adaptability and genericity, generally not covering the mission, safety and control loop aspects in a joint way. In addition, the lack of compatibility analysis between the recovery mechanisms implemented on different levels of the robot can have consequences on the safety and performance of the robot.This work describes a framework to limit the occurrence or propagation of faults in mobile robots in order to minimize the undesirable consequences both at the mission and safety levels. We propose an original multi-level approach dealing with 3 categories of undesired events. The first 2 categories concern 'high-level' events: violations of mission and safety constraints. The last category concerns 'low-level' events: faults, errors and failures occurring in the robot.In a preliminary phase, our approach requires the construction of fault trees for each of the previously identified categories. The study which is then carried out leads to a reinforcement of the reliability of the system in case of faults, by implementing local recoveries which act without disrupting the course of the mission. During the running phase, the undesired events are checked thanks to a model extracted from the fault trees. The system can then use local redundancy or recovery to limit the propagation of an undesired event or to suppress it. If an event is detected without a local fallback solution, a global recovery aiming at changing task is then proposed by a mission manager. The contributions are a complete framework to implement all the steps of this analysis, as well as an algorithm to recover from the events identified in the fault trees.We tested our framework on an underwater robot capable of performing autonomous missions. The tests were carried out within the framework of observation missions of the underwater ecosystem called transect. They demonstrate that our approach allows to limit the occurrence of catastrophic events while allowing an adequate management of the mission.