AI for Walking Robots

Robots didn’t just walk for us during the second seminar of Hilary Term’s AI for Good series at Reuben College – they danced!

Ioannis Havoutis’s talk was both educational and entertaining, providing an overview of the historical context for his research and the key concepts of AI and machine learning for walking robots. We gained insights into how robots learn to walk, and the complexities of mapping, reconstructing and navigating different environments. Ultimately, this means that undesirable jobs (the 4 ‘Ds’ of dull, dirty, dangerous, and dear) can be done by robots rather than humans.

The entertainment was provided by video footage of dancing robots from Boston Dynamics. The audience had a mixed response to the amusing choreography – some were enchanted by the trotting quadrupeds, others were unsettled by the implications of the steely humanoids. The uncanny facial expressions afforded by ‘soft’ robotics transported us to the world of science fiction; a rich source of human hopes and fears about the capabilities of robots.

Despite the brilliance of Ioannis’s talk, he was in some ways just the warm-up for the headline act: a real live robot demonstration. Our anticipation built as the quadruped was readied and we were given a tour of the robot’s anatomy as it emitted various noises and flashes, before rising up on powerful legs and stalking around the stage. We also got to see the ‘robot’s eye view’, the digital map of its immediate environment which allowed the robot to navigate, and avoid walking into Ioannis.

Robots then met dinosaurs in the Natural History Museum, where discussion over dinner shifted from the science of machine learning and motion to the ethics and utility of this branch of research. We tried to imagine using robots to care for our older relatives, or indeed our older selves. We acknowledged that robots could, in theory, be preferable to human helpers for some aspects of care as we age, but we were concerned about the potential loss of human contact that might follow. However, the limitations of robots made this rather a moot point – robots simply don’t have the capability at present to help with everyday care tasks, and it seems unlikely that there will be rapid developments in this field any time soon with a general lack of investment in care.

Thinking about robots doing care work got us thinking about the nature of work, how we value (or fail to value) different kinds of work, and the connection between these questions and the implication of robots on jobs in the global south. We were not persuaded that robots could liberate people from those dull jobs, in part because of the limitations of the technology (can a robot identify and pick ripe fruit like a human?) but also because of a healthy disregard for technological determinism – the assumption that technology itself can solve problems and drive progress. Instead, we tended towards a more socially determined view, exemplified in our pessimism about the military application of robots. Discussion that started with AI for walking robots broadened to a more general scientific and ethical question: can we ever really justify focusing on the science without thinking about the implications of the knowledge we are developing?


Gemma Hughes is a Research Fellow at Reuben College and a researcher in the Interdisciplinary Research in Health Sciences Group at the Nuffield Department of Primary Care Health Sciences.