Commande d'un robot collaboratif redondant en interaction avec des humains dans un contexte de manipulation et d'assemblage
|Abstract:||This thesis presents two novel control architectures for physical human-robot interactions (pHRIs) which are specically designed for the assembly industry. Indeed, two types of pHRI manipulators, each adapted to different industrial constraints and with different physical interaction interfaces, are studied each with their own control architecture. The rst pHRI manipulator designed is fully actuated and allows pHRIs in its free space, i.e., unilateral interactions, as well as pHRIs when its motion is constrained by the environment, i.e., bilateral interactions. The human force input can be applied on any of the manipulator's links because of the torque sensors in the robot joints. However, if a human force amplication is desired on the environment, then it is required to use the additional force sensor appended to the robot. Using this approach, combined with the signal of the force sensor at the end effector, it is then possible to use the ratio between the human and environment forces in order to generate the desired amplication. This control law is based on the concept of variable admittance control which has already demonstrated its great benets for unilateral interactions. Here, this concept is extended to bilateral interactions in order to obtain a single control algorithm for both states. A continuous transition can thus be implemented between both interaction modes which require different parameter values in order to achieve their optimal performance. The workow and results to achieve this rst control architecture are presented in three steps. Firstly, the control law is implemented on a single-degree-of-freedom (dof) prototype in order to test the amplication and transition potential, as well as the stability of the interaction. Secondly, a control optimisation algorithm is developed for bilateral interactions with a multidof robot. This algorithm assesses the system's robust stability using the structured singular value approach (-analysis), to afterwards, optimize the stable controllers in relation to a manipulator's conguration-dependent variable. This approach leads to a variable control law yielding a robustly stable system that can reach optimal performances for any robot conguration. In fact, the admittance regulator parameters follow a gain scheduling paradigm for bilateral interactions. The stability and performance of the system are assessed using impact tests on different environments. Finally, the optimal variable admittance control law is implemented and validated on a multi-dof robot (Kuka LWR 4) using different trajectory v tracking tasks for unilateral and bilateral interactions. The second pHRI manipulator designed is partially active and partially passive. The robot's mechanical architecture is known as a macro-mini. All actuated dofs which are part of the macro manipulator are doubled with passive joints which are part of the mini manipulator. This robot is therefore underactuated. The human operator interacts solely with the mini manipulator and, thereby, solely with the passive joints which leads to an interaction dynamics free of any delay. It is possible with this pHRI manipulator to dene a control law that yields an extremely low interaction impedance, even for heavy payloads. Despite the fact that force amplication is impractical with this kind of mechanism, bilateral interactions are stable for all sorts of contact. Moreover, the robot's cooperative and autonomous modes use similar control parameter values which enables an imperceptible transition from one mode to the other. The new control law is compared on different aspects with the previously-dened variable admittance control law. Results show that this new control law combined with the active-passive macro-mini manipulator, also known as uMan, leads to intuitive and safe interactions that are considerably superior to any interaction using a fully actuated manipulator. Furthermore, for the autonomous mode, an advanced collision detection and a specicallyadapted trajectory planning are developed. Experimental validations are presented in order to assess the ease of ne manipulation, to demonstrate the system's safety, and to establish the viability of the concept for the industry.|
|Document Type:||Thèse de doctorat|
|Open Access Date:||4 June 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.