27 May 2022,
by Theo Stouraitis, University of Edinburgh
My name is Theo, I am a research associate at the Statistical Machine Learning and Motor Control (SLMC) group at the University of Edinburgh (UoE). At the beginning of April, I returned back to Scotland, UK after a three-month research visit at the Interactive Robotics Group (IRG), at the Massachusetts Institute of Technology (MIT), Cambridge, MA, USA.
There, I worked together with Prof. Julie A. Shah who leads the IRG and Shen Li, a PhD student at IRG, on a joint project. Our research is focused on modelling the uncertainty of humans’ behaviours when performing a task in collaboration with another agent, such as a robot. Modelling uncertain human behaviour allows robots to decide when to move conservatively and when efficiently such that the human partner is always safe. To capture the uncertainty, we utilize mathematical objects that are called sets and we developed uncertainty-aware algorithms for estimation and prediction of the human behaviour that enable safe robot motion generation.
As an example, let’s consider a robot assisting a human to get dressed. As the cloth might occlude parts of the human body, like the human elbow, the robot needs to rely on alternate sensory inputs, such as sensed forces, towards estimating the posture of the human arm. Such estimates are likely to be erroneous to an extent, thus a robot should be able to evaluate the confidence that it has in these potentially inaccurate estimates of the human state to act accordingly and with caution.
In terms of outcomes, we are currently working on finalising a method that allows us to simultaneously estimate the state of the human and plan a safe path in continuous domains with constraints. Most of the work done at MIT was focused on composing the theoretical foundations of the method. Currently, we are continuing our collaboration remotely to produce experimental results on an assistive dressing task using our newly developed method and we aim to submit this work to a conference or a journal in the coming months.
In addition to this outcome, a collaborative paper that resulted from our remote collaboration prior to my visit at MIT has just recently been published. This is: S. Li, T. Stouraitis, M. Gienger, S. Vijayakumar and J. A. Shah, “Set-Based State Estimation With Probabilistic Consistency Guarantee Under Epistemic Uncertainty,” in IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 5958-5965, July 2022, doi: \doi{10.1109/LRA.2022.3154802} and its primary focus is on estimation with guarantees. This work was also featured at MIT news.
During my three-month visit at MIT, I had the opportunity to work and interact in person with all the members of the Interactive Robotics Group (IRG), as well as to participate in a number of events, where I met a wide range of academics from MIT and from other US and Canada institutions. I also had the opportunity to present the research work that we (my colleagues, my advisors and I) did in Scotland (at UoE) during my PhD at few other labs at MIT, USA and at the University of Toronto, Canada. All these interactions allowed me to grow my mindset and my network. The latter will hopefully lead to formation of new joint projects and joint events in the future. Hence, I found that my visit has been extremely stimulating and productive and I would definitely recommend it to any researcher. I believe that visiting another lab in another continent offers many new experiences and interactions that can serve as idea incubators.
Last but not least, I also had a lot of fun engaging in various MIT social activities and met a lot great people from all over the world.
Finally, I would like to thank SICSA for funding this research visit via the SICSA Postdoctoral and Early Career Researcher Exchanges (PECE) award, which greatly supported me during this fruitful research visit!