Georgia Institute of Technology
My research focuses on enabling in-home assistive robots to be more proactive so that they can assist users ahead of time and without being asked. Such proactivity can help alleviate the user’s burden of specifying detailed goals while also allowing the robot to plan time-consuming actions in advance. However such proactivity is challenging because it requires the robot to anticipate the user's actions and needs. I seek to enable proactivity over longer time horizons (minutes to hours) and spanning large spaces, such as entire households. I formulated this problem as longitudinal proactive assistance and attempt to build systems that address this problem through my research. Longitudinal proactive assistance requires accounting for needs beyond the user’s current activity, and has the added challenge of understanding their user's daily routines as a whole.
I aim to create systems that can learn human routines grounded in the physical environment, using unobtrusive observations of the user. By co-inhabiting the space with the user, the robot already has access to information about the user’s daily routines through observations of how the environment evolves in time and user’s actions (when visible). I seek to build systems that can leverage such information to develop an understanding of user’s routines such that it can predict their future needs. Additionally, I incorporate elements of explainability and active feedback so that the robot can interact with the human user to explain its predictions and seek feedback to clarify what the user intends to do, so that it can assist them more effectively.
To assist users through their daily lives, general purpose assistive robots will need to understand their user’s needs and not wait on specific commands, especially as their range of capabilities grows. They will need to adapt to the users’ changing needs and interact with them to improve over time. My research attempts to get one step closer to this ideal of a robot assistant that can develop a dynamic understanding of its user in context of their environment, improve it over time through observations and interact with the user, all towards assisting them through their daily lives efficiently and organically.