Predictive control for sensor-based robot tasks

Industrial robot manipulators play a central role in factory automation. A key factor in their success is that a single type of robot can be programmed for a variety of tasks. Previous research have developed a constraint-based task specification methodology to program these robots. This has proved to be a very powerful approach, even in complex environments where sensor-based control is required. However, current implementations of the methodology are instantaneous in nature and hence, purely reactive. This results in a greedy approach where the controller only does what is optimal for the current time-step, and does not consider how the task may evolve over time. The aim of this PhD project is to extend the instantaneous approach to a predictive strategy where the controller proactively plans future control actions ahead of time. This will increase the ability of the robot to better execute tasks while respecting the robot’s position, velocity, and torque limits as well as constraints imposed by the environment.