Motion and grasp planning for robotic manipulators in unstructured environment
Robotics is considered one of the leading enabling factors of industry 4.0. The classical workspace architecture of a robotic application usually sees the industrial manipulator confined to a cell where all the elements are supposed to be well known, and no uncertainty is likely to arise. This use starkly contrasts with the ongoing trends emphasizing a high need for flexibility embodied by the definition of “Smart industrial robots”, which is the cornerstone of the new industrial paradigm.
Industrial manipulators can tackle several industrial tasks involving grasp planning, such as palletizing or machine tending. Still, current techniques for robotic grasping work only in static scenes, rely on perception from only a single viewpoint, and have extended processing times, making them unsuitable for interacting with an unstructured environment. This type of approach cannot be used for tasks where object positional uncertainties are present, such as bin picking or decluttering and request other kinds of tasks, such as part assemblies, to be purposefully planned to avoid any deviation from nominal conditions, from which the system is unable to recover.
This research aims to study and develop control algorithms that allow the robot to move in an unstructured environment and interact with objects within it. Both classic probabilistic methods, as well as optimization and learning-based approaches, can be leveraged to enable the robotic system to react to dynamic changes in the surrounding. Finally, grasp synthesis techniques allow the manipulator to interact with the environment, enhancing the flexibility of the robot. They can also enhance robustness in structured or semi-structured tasks, accounting for unforeseen deviation from the nominal task condition that would lead the robot to halt its operation, reducing the design constraints on the robot station.
Back to Current Students