Assurance of Online Learning for Robotic and Autonomous Systems

Imagine autonomous underwater vehicles monitoring oceanic pollution over long periods of time to help address climate change, long-term robotic assistance enabling the elderly to live independently, or unmanned space missions acquiring essential scientific information with minimal ground-control interaction. The robotic and autonomous systems (RAS) from these applications share a common characteristic: they require safe online learning capabilities to carry out their critical missions.

Today’s RAS lack these capabilities. To compensate, they aim to handle uncertainty in their operating environment by employing multiple models (architecture, energy consumption, physical space, perception, etc.) built or learnt from data available at design time. In operation, these models can become invalid over time, and such brittleness may cause the system to deviate from its design intent, leading to safety violations, degraded functionality, and other undesirable effects. Enhancing RAS with online machine learning (ML) capabilities can mitigate this problem by ensuring that ML models are updated based on new observations obtained as the system operates. However, ensuring the safety of RAS online learning is very challenging because it requires the continual verification of evolving ML models.

This PhD project will devise continual verification algorithms and software tools that tackle this challenge. These novel algorithms and tools will generate assurance evidence establishing the safety of RAS that employ online ML techniques such as online reinforcement learning for decision making, online regression learning of resource usage and timing aspects of RAS, and online semi-supervised learning for environment predictive models.

Research supervision

The successful candidate will conduct your research under the supervision of Dr. Javier Camara (http://www.javicamara.com) and Dr. Radu Calinescu (https://www-users.cs.york.ac.uk/raduc/).

Apply for this studentship

1. Apply to study
• You must apply online for a full-time PhD in Computer Science
(https://www.cs.york.ac.uk/postgraduate/research-degrees/phd/#tab-4)
• You must quote the project title (Assurance of Online Learning-enabled Robotic and Autonomous Systems) in your application.
• There is no need to write a full formal research proposal (2,000-3,000 words) in your application to study as this studentship is for a specific project.

2. Provide a personal statement. As part of your application please provide a personal statement of 500-1,000 words with your initial thoughts on the research topic.

We will look favourably on applicants that can demonstrate knowledge of machine learning and/or formal verification/model checking techniques.