Current robots in manufacturing settings are not able to adapt to variations introduced by collaborating humans or other factors in the environment. This is because there is still a need for robots to be equipped with the appropriate cognitive architecture and approach.
The real time cognitive ability of humans and animals provide us a lot of inspiration for creating resilient and robust artificial intelligent systems that optimise in real time to problems and variations in their environment. Achieving human inspired cognition on robots would enable them to be easily taught new skills (reprogrammable) and flexibly redeployed onto various tasks in manufacturing and other settings.
One of the paradigms that has been created by the machine learning community is the Reinforcement Learning paradigm. In reinforcement learning, the focus is on applying rewards from the environment to learn as well as finding a balance between exploration (of uncharted territory) and exploitation (of current knowledge) to find solutions to problems. The paradigm relies on a prior well-defined environment and results in a control policy that is best suited for the environment that it was designed for. It starts suffering when there are variations in the environment.
Informed by cognitive psychology literature and supported by neuro-evolution , meta-learning and neural architecture search methodologies , this PhD topic aims to develop new real time cognitive approaches and computational to create robots that can flexibly learn new tasks, transfer learnt tasks to new settings and adapt in real time to variations in the environment.
- Investigate various cognitive psychology frameworks that supports how humans use previously learnt skills to deal with variations in the environment
- Develop a cognitive framework that embeds meta-learning and neural architecture search methodologies
- Develop a robotics simulation environment for various industrial relevant use cases. You will have an opportunity to collaborate with our industrial partners on various use cases.
- Integrate objectives 2 and 3 in the development of new computational intelligent algorithms.
- Validate algorithms on 2 physical collaborative robot platforms. You will work on various robot platforms including the Franka arm, Sawyer, UR5 and UR15.
This work will be carried out in Department of Computer Science (top 10 in UK for research quality) YorRobots (https://www.york.ac.uk/yorrobots/) and at the University of York’s £15m Institute for Safe Autonomy (https://www.york.ac.uk/safe-autonomy/). The Institute for Safe Autonomy is UK’s first research centre dedicated to the design, development, safety and communications for robotics and connected autonomous systems. The Institute provides a world-leading ecosystem for research and innovation, education, public engagement and commercial realisation.
The University of York is part of the research-intensive Russell Group of Universities in the United Kingdom which inject nearly £87 billion into the national economy every year. We believe that people and ideas are the key to meeting global challenges. Through world-class research and education we are helping to create a dynamic economy, stronger communities and a better future for the world. We maintain the very best research, an outstanding teaching and learning experience and unrivalled links with local and national business and the public sector.
Russell Group universities have huge social, economic and cultural impacts locally, across the UK and around the globe. The Russell Group of Universities produce more than two-thirds of the world-leading research produced in UK universities and support more than 260,000 jobs across the country. 32% of students are of non-UK nationality, attracted to our universities by the quality, relevance and reputation of research.
Russell Group members have a strong role and influence within their regional and local communities, collaborate with businesses on joint research projects and supply highly-qualified and highly-motivated graduates to the local workforce.