Mobile Robotic Inspector: Learning to Explore and Manipulate in the Real World

This collaborative research proposal between the Oxford Robotics Institute (ORI) and the ETH extends themes of embodied robotic intelligence outlined in Oxford's ongoing Programme Grant "Embodied Intelligence: From Sensing to Collaboration" to include physical intelligence - specifically dynamic manipulation in contact-rich environments and perception suitable for autonomy in wild, natural environments.

We propose a strategic collaboration with Prof. Dr. Hutter and Dr. Cadena of ETH Zurich who are world leaders in dynamic locomotion, manipulation and navigation and ETH's Center for Robotics - a coordinated ETH initiative to grow robotics in ETH and Switzerland. Their Robot Systems Lab (RSL) regularly receives plaudits and recognition and has created 6 spin-outs. Interaction with other academics within the Center (e.g. soft robotics, autonomy, controls) is also envisaged.

ORI brings expertise in perception for state estimation and navigation, deep learning for locomotion and manipulation and decision making, while ETH has a strong track record in mobile robotics, dynamic locomotion and control. Together, we will develop new approaches to address the challenges of autonomous manipulation and navigation with dynamic robots.

Our proposed project is organised into two research strands, which will use the ANYmal quadruped robot as a common platform. The first strand will focus on 'Data-Driven Perception and Navigation ', while the second strand will focus on 'Contact Rich Manipulation for Mobile Platforms'. Together with ETH we have designed an exciting exchange programme that will strengthen our collaboration and benefit from substantive industry support.

In addition, we plan to host a standalone symposium to promote the common research links between Oxford and ETH, and another public outreach event linked with one of the main robotics conferences (IEEE ICRA, RSS).

In summary, the proposed collaboration presents a unique opportunity to collaboratively explore the challenges of unstructured autonomy and contact-rich manipulation by pairing advances from the first half of our programme grant with expertise from ETH's world class faculty and laboratories. Our proposed collaboration is timely, given the growing

importance of the proposed research, while our unique combination of expertise has the potential to have significant impact both in UK academia and in a range of business sectors. We propose a clear research and demonstration plan involving collaborations with several companies in the nuclear sector and the manufacturer of a leading quadruped robot. The project will conclude with live on-site trials at a real nuclear facility.