Data-driven learning and control for the development of safe Collaborative Robots (Cobots)

Machine learning has led industry a long way on the road to autonomous intelligent robotics. However, the high safety standards required for the deployment of safe collaborative robotics call for control methods with theoretical safety guarantees. At the same time, theoretical formulations are often only as good as our ability to solve the associated problems fast and efficiently on embedded devices with limited resources. This project will aim at developing a novel control theoretic framework that will fuse together model predictive control and statistical learning theory, and optimisation algorithms that will be amenable to massive parallelisation on GPUs and will enable their deployment on next-generation collaborative robots (known as cobots). We envision the development of context-aware cobots that while the accumulate data, they will be able to become “more confident” and respond in a better, more natural yet safe manner to the motion of their human collaborators.

Project Description:

The main pillars of this project will be:

1. Data-Driven Model predictive control: Model predictive control (MPC) is an advanced optimisation-based control methodology that can take into account nonlinear dynamics and input/state constraints. Risk-averse MPC is a novel flavour of MPC that can guarantee safe operation under high-effect low-probability events (HELP), such as those associated with unexpected human motion. The development of truly context-aware collaborative robotics necessitates an amalgamation of statistical learning and control theory that fits perfectly into the risk-averse control framework. Moreover, we will formulate and study the problem of human-robot interactions as a game and will leverage relevant results from game theory.
2. Fast Numerical Optimisation: The benefits from the use of MPC in robotics have long become evident. Yet, its industrial uptake has been hampered by the availability of reliable and efficient numerical methods to solve the associated optimisation problems in real time on resource-constrained devices. In this project we will propose novel numerical methods for nonlinear MPC with provably superior performance compared to popular SQP/IP methods.
3. Embedded High-Performance Computing: we will harness the computational power of embedded hardware equipped with GPUs (e.g., NVIDIA Jetson) and the structure of the optimisation problems involved in MPC to enable the real-time solution of complex, nonconvex, large-scale optimisation problems such as those arising from risk averse MPC formulations.

The final outcome will be the development of autonomous humanoid robots that will be able to perform collaborative tasks with safety guarantees. The robots will use data, obtained from vision-based systems and other sensors, to build and update motion models and interaction models with the humans in its vicinity. The more data the robot gathers, the more confident it will be about moving safely in the shared workspace.

Project Key Words: Cobots; Data-driven control; Model Predictive Control; Fast Numerical Optimisation; GPUs.

Start Date: 01/10/22

Application Closing date: 28/02/22

For further information about eligibility criteria please refer to the DfE Postgraduate Studentship Terms and Conditions 2021-22 at https://go.qub.ac.uk/dfeterms

Applicants should apply electronically through the Queen’s online application portal at: https://dap.qub.ac.uk/portal/

Academic Requirements:

A minimum 2.1 honours degree or equivalent in Computer Science or Electrical and Electronic Engineering or relevant degree is required.

Funding Notes:

This three year studentship, for full-time PhD study, is potentially funded by the Department for the Economy (DfE) and commences on 1 October 2022. For UK domiciled students the value of an award includes the cost of approved tuition fees as well as maintenance support (Fees £4,500 pa and Stipend rate £15,609 pa - 2022-23 rates to be confirmed). To be considered eligible for a full DfE studentship award you must have been ordinarily resident in the United Kingdom for the full three year period before the first day of the first academic year of the course.

For candidates who do not meet the above residency requirements, a small number of international studentships may be available from the School. These are expected to be highly competitive, and a selection process will determine the strongest candidates across a range of School projects, who may then be offered funding for their chosen project.