The aim of my PhD project was to endow the Anthropomorphic Robot System ARoS with human-like reaching, grasping and manipulation, by combining diverse skills in Engineering, Mathematics and Human Motor Control. This robotic platform is used in human-robot collaboration tasks and was designed and built at the Department of Industrial Electronics, University of Minho.
ARoS consists of a static torso, equipped with a seven degrees of freedom anthropomorphic arm, a four degrees of freedom three-fingered hand and a stereo vision system mounted on a pan-tilt unit.
Due to the redundant degrees of freedom of its robotics arm and hand, a goal directed action, in everyday tasks such as reach-to-grasp an object, may be achieved in multiple ways.
Using knowledge from experimental work and functional models of human upper limb we developed a model for movement planning and control of ARoS' robotics arm and hand for reaching and grasping objects in three-dimensional space for different objects and grip types, and manipulate them. Our aim was to generate smooth, fluent and collision-free arm and hand movements that qualitatively explain the main characteristics of hand and arm trajectories observed in experiments with humans. These characteristics together with an anthropomorphic shape have been pointed out as key characteristics for facilitating human-robot collaboration.
We were strongly inspired by the posture-based model of Rosenbaum and colleagues and were particularly interested in exploiting the obstacle avoidance mechanism proposed by them for the robotics domain.
The movement planning and control model that was developed is used as part of the robot control architecture to generate overt motor behavior in different shared tasks. We have tested the movement planning for simple reach-to-grasp movements but also in more complex tasks such as the construction of a toy vehicle or assisting a human to drink, which involve manipulation of objects and communicative gestures.
The results show that our strategy to borrow principles from Human Motor Control is a feasible approach towards more human-like movements of an anthropomorphic robotics arm and hand.