Collaborative Control Strategies for Robotic Manipulation

Since the dawn of the Industrial Revolution, humans have been striving to increase productivity through the use of automatic machinery such as robots. With the Third Industrial Revolution, robots saw widespread adoption in the world of manufacturing. In the strives for efficiency, they offered a solution to mundane or repetitive human tasks. Now with Industry 4.0, where general purpose and smart robotics are its two main themes, the question is how best to exploit this technology? This thesis presents novel work which attempts to answer these callings. With recent improvements in technology advancing the field of robotics, such machines are no longer bound to a single task. Robots come with a suite of sensors, allowing them to take into consideration their surroundings. Ever increasing processing power and improved algorithms enable them to accomplish more complex or uncertain tasks. Additionally, it has become possible to detect and respond to events in real-time. For fully autonomous robots, to safely and successfully operate in unstructured environments, further advances and improvements in the field of robotics must be made. Slip detection and control is an area which can be applied to dual collaborating manipulators. With adequate sensing capabilities, two collaborating manipulators may hold an object without the use of dexterous end-effectors. This vastly improves the versatility of these robots as dual arm or multiple collaborating robots may lift heavy or awkward payloads that a single robot could not. The fast control required to prevent slip lead onto the next area of research using model predictive control (MPC). Whilst MPC saw its first use in the 1970s in the chemical industry, the use of MPC on fast, dynamic robotic systems is still an area requiring open to improvements, especially in terms of app the area of dynamic obstacle avoidance. Initially MPC for a simple point to point problem is investigated on a single robotic manipulator. This approach combines path planning and tracking for robotic manipulators, as opposed to the normal approach of externally provided paths with MPC used only for tracking. The same approach is then applied to dual and triple manipulators, both in two and three dimensions. This is believed to be the first time MPC has been applied to a manipulator synchronised motion task. The area of MPC obstacle avoidance was next investigated. Working in shared environments, robots must have the capability to avoid other entities and obstructions, so this approach improves the usefulness of a robot equipped with such a controller. Again, no externally supplied path is required. This is successfully applied to both a two and three dimensional single end-effector. Deterministic obstacle position prediction is also taken into account for the MPC problem which greatly improves the efficiency of motion and comprises another novel application. In the real world, most scenarios are non-deterministic and susceptible to agents undergoing random motion. Following the successful implementation of MPC obstacle avoidance using constraints, the next step was to adapt the system to solve such a problem. Using probabilistic based motions, this can be achieved through stochastic MPC (SMPC). A scenario tree is used to represent the stochastic motion of an obstacle. From this, probabilities are gathered and considered as part of the MPC problem. Through simulations, the SMPC approach is shown to greatly reduce the chance of obstacle collision over the standard MPC. This work is also believed to be the first use of SMPC on a robotic manipulator, further improving the practical applicability of MPC on these type of systems.