Our aim is to develop biologically inspired, general-purpose reasoning, planning, and control algorithms for physical, compliant, and safe human-robot collaboration in real, dynamic environments. This research direction is coined as Collaborative Planning and Control, where human biomechanical and cognitive behavior models are taken explicitly into account for decision-making and control. It encompasses several short- and long-term challenges, such as: (i) Learning multimodal representations from imperfect/noisy sensory information. (ii) Modeling human behaviors, their cognitive skills, and interactive roles in the loop from a handful of available data. (iii) Real-time, scalable decision-making using multimodal sensory information and human behavior model in the loop under computational and safety constraints.
Our general-purpose and human-aware reasoning, planning, and control algorithms touch a variety of applications, including but not limited to: 1) Healthcare & Industrial sectors where autonomous mobile manipulators at home, in the hospitals, or on the factory floors can enhance the workforce on multiple fronts, from maintaining the supply and delivery chain to performing complex decision-making and control tasks such as non-invasive surgery. 2) Search & Rescue in which dynamical autonomous team formation under hostile environments is needed for rapid area coverage, surveillance, manipulation, and other services for the risk-free accomplishment the given tasks. 3) Smart Farms, with Purdue being a land-grant university, our work naturally paves its way to agriculture sectors solving various complex tasks such as deformable object manipulation and handling, pest control, and human-robot collaboration for a wide range of agricultural tasks from plowing to harvesting. 4) Autonomous driving, where human-robot interaction is inevitable, would require shared autonomy and fluid transition of control between involved entities.