• planning
• grasping
• manipulation
• form-closure
• force-closure
• optimization
• rrt
• ros
• moveit
• franka

This thesis addresses the problem of planning robotic manipulation for objects tightly constrained by the environment. In most cases, state-of-the-art planners struggle to fit satisfactorily in low dimensional sub-manifolds, while still ensuring geometric and force feasibility. On the other hand, humans are at ease with such situations and indeed exploit constraints to manipulate objects proficiently. A promising state-of-the-art MATLAB algorithm aims to elevate automatic grasping performance up to the level of humans by merging randomized grasp planning methods with model-based grasp analysis. By using partial form-closure analysis, the algorithm finds the set of geometrically feasible motions of box-shaped objects, and with force-closure analysis ensures that such motions are physically feasible. In my work, I created a planning and simulation environment using the C++ and Python languages, that let the algorithm run natively in a ROS environment for prompt interfacing to real robotic systems. The proposed software features integration with the Graphical User Interface of MoveIt for ease of visualization and debugging. The system is modular with respect to different solver components that can be used both in the form-closure and force-closure components, to optimize its performance at run-time. Moreover, I worked toward extending the algorithm to be able to plan for objects of any shape immersed in an environment of custom complexity, removing the previous limitation of dealing only with box-shaped items. A preliminary version of this ROS implementation has been extensively tested in simulation to analyze and break-down the computational cost of the different components of the planner in various meaningful scenarios, comparing it to the previous MATLAB implementation. Finally, the algorithm has been demonstrated in preliminary experiments with a collaborative robot, showing that the implemented framework features the ability to compute solutions for heavily constrained real-world manipulation problems.