Control of robots in a physical collaboration. Challenges, methods and applications

Traditional robotics, particularly in the industrial sector, is based on planned robotics, where the robot's actions and movements are predefined according to a highly structured and controlled environment. Because of this dependence on a rigid environment, these robots are often isolated from humans and other dynamic systems, in order to avoid any unforeseen events, such as unexpected modifications to the environment (changes in the layout of objects or equipment, and the appearance of new obstacles).

These robots are largely position-controlled, where a trajectory is planned beforehand, and then followed using an adapted closed-loop control law. There are also classic robotics applications, where control is based on effort, but on a static or little-changing environment, and where an effort setpoint is defined beforehand.

In recent years, we've seen the arrival of a new generation of so-called "collaborative" robots (commonly known as "cobots"), which are winning over more and more manufacturers and boosting the interest in robotics in certain fields by making them relatively easy to use and reprogram.

Although these cobots are often used like conventional robots in terms of application, with higher human safety and collision detection capabilities, real cobotics (collaborative robotics) needs to be rethought in terms of control to make robots collaborate more with humans. In this case, the approach of pre-planning the robot's trajectory becomes inadequate, as the unpredictable physical interaction of the human, which serves to guide the collaboration, can no longer guarantee the correct execution of this trajectory. In this context, where the levels of physical interaction can be varied, different robotic control methods have to be found and applied to the context of the physical interaction that is posed.

This article focuses particularly on the case of continuous physical interaction between humans and robots, where both are active at the same time, to accomplish common tasks, such as comanipulation or object cotransport. The article deals with the control aspects of robots in this context, where so-called "compliant" methods are described and adapted to the context posed, and exploit the robot's dynamic model. In addition, problems are addressed and solutions provided, linked to the need to adapt the level of on-line compliance and to the presence of a comanipulated object whose dynamics present uncertainties. Thus, the challenges posed in this context will be outlined in section 1