This is an opportunity to work in a new area of robotic control and movement. The impact of which is applicable across a wide field of manufacturing, wearable prosthetics or testing scenarios.
The Control Systems Research Group, with which this position is based, has a well-established research expertise in the application of advanced control and condition monitoring concepts for a variety of industrial applications. In particular the group has an internationally-leading position in the application of these concepts for railways, but also has expertise in other areas such as aerospace and energy systems. This position looks to expand on our current expertise into new areas of research.
Control of Robotic motion is highly dependent on the position control of rigid linkages around various joints and design. Although high power, precision robots exist, they typically operate by tolerating the flex and bending modes of the rigid linkages.
This project is concerned with optimising robotic movement by making use of (or designing in) deliberate bending and spring dynamics into systems to make more efficient movements. This is more akin to applications in biology where natural bending and flexes can be used within a body for more efficiency. This is extended to application scenarios such as; ball throw, golf club swing, or any 'springing' motion seen in running/jumping, where optimum efficiency is obtained by storing energy in shaft bending before allowing this to ‘unload’ at the point of launch. Typical robotic systems struggle to achieve this optimum motion due to the position-focussed nature of the control schemes and lack of system wide optimisation.
Traditional robotic movement systems rely on rigid body connections between actuation points or joint couplings. For applications requiring compliance and control of contact forces, impedance control allows the force/position relationship to be specified at an end effector, but structural components and joints are still rigidly constrained. In such robots, structural flex is detrimental to performance and complicates or invalidates traditional forward and inverse kinematic approaches. Yet in the natural world, muscle/bone structure rely on elasticity combined with actuation to optimise energy use in ways that robots struggle to mimic
This provides a more lightweight structure and allows lower control forces and torques to be used, but complicates control design and decreases robustness to uncertainty in, for example, the mass of a ball or golf club.
Accurate high-performance control of such robots requires detailed consideration of the elasticity and bending dynamics of the manipulator, and controllers that account for these nonlinear dynamics and adapt rapidly to uncertainties. This project aims to develop these controllers, drawing on recent advances in adaptive and robust nonlinear MPC and reinforcement learning. Successful candidates will develop skills in multibody dynamics and predictive control applied to flexible robotics and autonomous systems.
Possible research questions in this study are:
- What control techniques are required for lightweight, flexible robots? (may not be novel)
- Can we take advantage of elasticity in certain types of robotic motion, especially those involving target velocities as well as positions? (thinking of throwing balls, etc)
- Can we make this control adaptive, e.g. to different masses of balls etc (in particular, there’s a big open research question about adaptive MPC)
- do we need to ‘probe’ the system in some way during control to update our models for adaptive control? (e.g., feeling the weight of a ball before you throw it)
- can we codesign the robotic manipulator with the controller for a specific task? (maybe a bit more ambitious/tenuous)
Applications of this could be in replicating sports movements for product testing, or in more efficient movement in wearable robotics and product manufacture.