Abstract
The growing need for flexibility in industrial settings leads to reconsidering the way robotic systems are exploited in such environments. It follows that the relationship between humans and machines has to evolve in favor of more proximity, by letting them share the same workspace and physically interact together.With this in mind, this thesis aims at contributing beyond the state of art in the control of dual-arm robots for collaborative purposes in an industrial context. We propose a generic online kinematic control approach based on an admittance control law which enables safe manipulation of objects in physical collaboration with humans. The controller solves a acrfull{qp} optimization problem to find the joint space motion that satisfies the task space command while respecting a set of constraints (e.g. joint limits, collision avoidance).The kinematic solver can be tuned to generate parsimonious solutions at the joint velocity level, meaning that as few actuators as possible are activated to achieve the tasks. This induces potentially safer behavior in an unstructured environment shared with humans.Dual-arm platforms are sometimes extended to include additional robots (e.g., mobile base, articulated torso, ...). In this thesis, we also present an original hierarchical method for the control of multi-robot systems.An open-source implementation of the work, acrfull{rkcl}, is available. It implements all the components described in this thesis and can be easily configured to work with new robots. Throughout the developments, experimental validations have been performed on the dual-arm mobile cobot BAZAR.