Our research is focussed on learning, adaptation and control mechanisms needed to improve the skills of autonomous robots, improving their ability to cope with unknown and dynamic environments. Even as we work towards enhancing the autonomy of robots, we want these robots to be intelligent enough to be able to effectively collaborate with human co-workers (to learn from them, to take instructions from them and to coordinate joint actions with them). With the rapid development of such robotic skills, it is becoming increasingly more important that autonomous systems be safe, trustworthy and responsibly developed.
Our approach is to couple AI methods development with the development of complete systems that can meaningfully address realistic applications. Salient application areas of interest include the following.
Dexterous Manipulation Skills: We are fascinated by the richness and expressivity of human/animal manipulation, ranging from the lightness of touch when handling soft materials in our daily activities, to the generality of our ability to plan with novel objects in uncertain environments. We would like to understand these behaviours well enough to be able to devise autonomous robots that can behave similarly. Such skills are needed in many domains, ranging from advanced manufacturing to surgical automation.
Physical Scene Understanding and Active Sensing: The combination of novel sensors mounted on or manipulated by robots, with physics-informed machine learning methods that enable efficient inference of causal, interpretable, multi-scale models, represents fascinating new opportunities for real-time diagnostic systems and their incorporation into closed-loop robotic systems. We are developing this technology with collaborators from medicine, and the bio-physical sciences.
Intention-aware Planning in Human-centred Environments: Prof Ramamoorthy’s work at FiveAI (with a diverse team including a few RAD alumni) has taken several ideas developed within our research group into a technology stack that has been deployed and demonstrated on public roads (e.g., in the StreetWise trials in London). This includes methods for prediction and hypothesis testing in multi-agent situations, and safe motion planning in interactive dynamic environments.
Funding for our research work has come from the following sources (selected grants, listed in reverse chronological order):