Decision and control functionalities for robots

Work package (WP) 3 is focused on the problems of decision and control of robots. Once the robot has perceptual information about its environment, it must plan the future actions to be executed to perform the task for which it is dedicated. In this domain, the state of the art is very rich, ranging from global motion planning techniques to sensor-based control. This WP is structured around 3 scientific challenges.

In the context of decision making, we will focus on decision making algorithms that propose solutions when robots do not perform repetitive tasks but plan their future actions for different tasks over time. Indeed, when a mobile robot is required to interact in the real world, it faces a double challenge during its decision making phase: (1) its movements must be safe, and (2) "acceptable" from the point of view of people present in the robot's environment. We will study these two challenges and implement solutions on the project's robotic platforms.

On the other hand, when executing actions, real-time control of robots is mandatory to make sure that potential collisions between the robot and its environment result in a low probability of serious injury even at speeds as high as 2 m.s−1. Collisions would better be avoided nevertheless, not only because of the remaining risk, however low it is, but also because each collision disrupts the tasks of the robot. To adequately tackle this issue, sophisticated numerical methods will be developed to couple the differential laws from multiple physical models. These models will be the support for control design analysis.

Finally, continuum robot control will be carefully studied as they evolve in fragile environments related to medical applications (human body, tissue, organs, etc.). Indeed, while continuum robots can be quickly prototyped and fabricated, and highlight diversity in motion generation along their body and stiffness variation, developing adequate controllers is still an open question in the community. Most controllers omit their inherent flexibility and infinite number of degrees of freedom, or at least do not fully take advantage of those. In contrast with control of rigid robots, not only end-effector control is necessary, but also the entire shape of the robot as it navigates within a highly constrained environment for safety purposes.