Exploring Self-Motion Manifolds to Optimize Task-Oriented Manipulability of Redundant Robots

As the field of robotics has a surge like never seen before, new possibilities for uses of robotic manipulators are surfacing every day. Development in assistive robotics is faced with new challenges, as the tasks become more complex and intricate. Every day acts like opening a door or pouring a glass of water, that are very simple and intuitive for humans, put systems like EDAN to test. Kinematic and physical constraints, e.g. joint limits and singularities, have to be actively avoided to guarantee the success of the task. Some approaches, like in [1], have shown great results in improving the manipulability of manipulators. However, these approaches have mainly a step-by-step, local approach to the problem. A self-motion manifold represents all the possible configuration of a manipulator for a certain Cartesian position. By studying them, it is possible to know beforehand what joint configuration will allow the manipulator to be as successful as possible in the task at hand. It allows the robot to rearrange itself in order to reach the global best, allowing to intentionally pass through worse configurations. This idea provides the advantage against step-by-step optimizations, as it can leave difficult situations, by having a global approach instead of a local one. The optimization was tested in simulation as well in real systems and it showed promise and improvements when compared to state-of-the-art methods