Planning and Control of a Collaborative Manipulator in Heavily Constrained Scenarios

The automation of the logistic processes of large e-commerce warehouses is one of the biggest challenges of today's robotics. The rise and development of a new generation of collaborative robots is now providing new ways to tackle this problem. Indeed, their use in pick-and-place operations is quickly spreading, saving time and energy, and preventing human errors. Nonetheless, several issues remain open, one of the most relevant is the automatization of bin-picking tasks. Many picking tasks for objects stored inside boxes are planned using hard-coded and custom solutions, which lack of flexibility and need human intervention to adapt to new objects and/or boxes, or even executed directly by human operators. The golden reference for dexterity and manipulation, the human wrist-hand system, still outperforms robots in such tasks, because of its adaptability and compliance. Taking inspiration from it, a recent solution to increase robot dexterity consists in embedding robotic wrists on top of manipulators. An example such as device is the Compact Soft Articulated Parallel Wrist proposed by Negrello et al. This solution can be used to address the problem of grasping in narrow spaces such as boxes. Notwithstanding the increased dexterity, autonomous planning for picking objects in boxes using the aforementioned device is still a challenging problem due to mainly two factors. Firstly, the heavy constraints imposed by the box walls create lower dimensional manifolds in the planning space and thus prevent direct applicability of state-of-the-art sampling-based motion planning techniques, which strive to fit into narrow spaces. Furthermore, the high number of degrees of freedom of manipulators mounting robotic wrists bring about the issue of redundancy which should be solved while accounting for robot kinematics. This thesis proposes a novel solution to autonomous planning of bin-picking operations. The solution exploits the increased dexterity and redundancy given by the soft wrist to design a redundancy resolution method able of naturally including the box-constraints. The method leverages the Reverse Priority redundancy resolution algorithm proposed by Flacco and De Luca with the inclusion of unilateral constraints on a limited set of points of the robotic structure to plan collision-free trajectories for the manipulator. The proposed method is validated through experiments of grasping inside a box typically used in warehouses, comparing the results with MoveIt! sampling-based algorithms on the system Franka Emika Panda, Compact Soft Articulated Parallel Wrist and a Pisa/IIT SoftHand as end-effector. The whole framework is implemented in ROS and Matlab/Simulink. By comparing with the traditional RRT-based motion planning, the advantages and limits of the proposed approach are shown and discussed