Control architectures for robotics - Approaches and trends

Robotics is inherently multidisciplinary, dealing from the outset with problems that combine mechanics, automation, electronics, computer science, etc., not forgetting those relating to the field of application itself. It owes its progress to the advances made in each discipline, as well as to the cross-fertilization between them. Robots are increasingly complex technological entities, entrusted with ever more complicated missions in a multitude of application fields. The specificities of the various fields of application and the main categories of robots (land, underwater and airborne robotics) are often at the root of different solutions. Some applications will require a high degree of reactivity and less decision-making capability, while others will require a high degree of planning, and so on.

This complexity places a number of demands on the IT system that is supposed to manage the robot's operation and support its ability to act, adapt, make decisions, etc., the "intelligence" conferred by its control. It is therefore the control software architecture, i.e. the way in which the software responsible for controlling the robot is designed and developed, that we will discuss in this article.

These software architectures are characterized by the need to take into account a whole spectrum of functions, from time- and space-intensive deliberative calculations to real-time control of physical devices. Organizing these "functions" and linking them requires a high level of expertise. This is why general solutions were initially proposed in the form of "architectural models". Classically accepted architectural models will be presented, bearing in mind that they are still subject to evolution to integrate the variety of relational schemes that give these architectures a unique dimension. Indeed, robots immersed in their environment are also subject to human-robot relations (shared decision-making autonomy), robot-robot relations (flotilla) or even robot-active environment relations (sensors in the home and intelligent roads, for example) which, when applied to control architectures, require consideration of the dynamic relations that can be established.

While the architectural model provides guidance on how to structure an architecture, it remains to define how to describe and implement it. With the growing complexity of robotic software architectures, and the explosion in the ever-greater diversity of applications and missions, the design and development of high-performance, correct software architectures is becoming a major challenge. To date, there is no commonly accepted approach to capitalizing on and pooling knowledge and best practices, or to reusing the software building blocks developed. The inventory we are about to draw up shows the...