• control
• environment
• grasping
• hands
• manipulation
• planning
• real-world
• robot

Humans display extraordinary ability in grasping and manipulating objects even in extremely complex scenarios. With the ambitious aim of reaching man’s ability, several robotic manipulators and hands were designed in the past decades, thus increasing the dexterity of robots considerably. Notwithstanding such advances in design, robots are still far behind humans in the ability to manipulate objects in real-world scenarios. A main reason is the lack of algorithms which take into account - at the different stages of planning and control - the heterogeneity of the robots, the constraints and the uncertainties permeating the unstructured external world.

Multi-robot systems for object handling are becoming increasingly popular in warehouses and factories, since they enable the development of more versatile, efficient and robust systems than single robots. However, this leads to increased complexity in planning and dispatching actions to robots. Furthermore, in many realistic situations, solving manipulation planning problems requires dealing with heavy and/or tight constraints imposed by the environment on the object. In such cases, existing planners struggle to fit satisfactorily in such low dimensional sub-manifolds. In addition, as a consequence of imperfect modeling or data-acquisition, the execution of computed plans may fail. As mechanical compliance aids only partially in coping with such execution issues, reactive strategies and hand control methods that take into account the uncertainties given by the unstructured environment are still a topic of research.

In such a context, this dissertation discusses the role of human-inspired solutions and of environment exploitation in robotic manipulation. This leads to introduce some model-based tools which address the aforementioned deficiencies at the three different levels of planning, reaction and control. Firstly, a task-planning algorithm for object handling - that is both scenario- and robot-independent - is devised; it tackles explicitly the problem of planning for multi-robot systems. Then, a random-tree-based motion planner is proposed: it can be used to solve manipulation problems with tight environment constraints. The output motion plans are constructed to ensure geometric and force feasibility even when the object is heavily constrained. Going to lower levels, a human-inspired reactive strategy for robotic hands, and a novel impedance control law for tendon-driven hands are presented. The former attacks the issue of grasp failure caused by object positioning error and the latter is aimed at providing a robust interaction control of the hand in the presence of unexpected collisions. In short, following a top-to-down approach, the objective is to provide solutions at different levels to further enable the adoption of robotic systems in real-world applications by going towards an environment-aware manipulation.

The developed concepts are validated using state-of-the-art robots in real-world scenarios. The evaluated tasks, ranging from complex grasping and manipulation to heavy interaction, demonstrate the capabilities of the formulated tools to effectively deal with the limitations imposed by realistic scenarios by turning such disadvantages into opportunities to be exploited.