• Obstacle Avoidance

Autonomous robotic grasping is still an open problem. Literature proposes several approaches, but they do not tackle all the involved challenges. These include (i) planning of the end-effector pose, and (ii) planning of a collision-free motion for the robot. A recent work proposed the algorithm “Grasp-it-like-a-Pro” to tackle the challenge (i). Here, the Authors propose an approach to transfer grasping expertise from an expert human operator to a robotic manipulator. The output of the algorithm is a pose for the end-effector of the robot that enables grasping objects of unknown and general shape. However, the problem of planning a collision-free motion was not covered. This means that the success of the grasp may be hindered by undesired collisions between the robot end-effector and the environment, including the object to-be-grasped.

In this Thesis, I propose a motion planner to solve the challenge (ii). In particular, I develop an algorithm that plans a collision-free motion given an initial robot position, a final desired pose and a description of the environmental obstacles. The method is based on a reverse task-priority formulation for the kinematic control, where the obstacle avoidance represents the task with the highest priority. I propose three methods for designing this task and the activation policies, presenting them in order of increasing complexity.

Finally, I provide simulations and experiments to validate the proposed techniques. The first two methods are tested only with simulations, while the third one is also tested on Franka Emika Panda manipulator equipped with a Pisa/IIT SoftHand.