Human-Centered Modeling and Control of Cooperative Manipulation with Bimanual Robots

Duration: 2014-2017
Funding: NSF NRI (#1427260 )
Collaborators: Oussama Khatib (AI Lab, Stanford University)
Researchers: Aaron Bestick, Robert Matthew, Ruzena Bajcsy, Gregorij Kurillo, Oussama Khatib (SU), Samir Menon (SU)


  • Enable improved control of robots providing direct physical assistance to humans
  • Create unified model of the human-robot coupled mechanical system
  • Predict intent of human operator based on physical cues

    This proposal addresses modeling and control aspects of human-robot interaction by considering constraints imposed by an individual's physiology. The project is motivated by increasing demand for automation in unstructured environments that require high-level cognitive processing and complex decision-making which cannot yet be fully automated. By taking human-centric approach, data-driven musculoskeletal models are incorporated into the robot interaction model to account for differences of individuals.

    Each cooperative activity is divided into action primitives requiring different control strategies while estimating human intent from various sensors. The framework is based on theory of hybrid systems that provides provable safety and stability criteria. The outcome of this research will facilitate methodology for safer and more reliable human-robot interaction and advance state-of-the-art in human movement analysis and control theory. The broader impacts of this research will be realized through new insights into understanding of human intent and haptic cooperation applicable to general human-machine interaction. With increasing interest in service robotics safe and reliable interaction will be the key to successful introduction of robots in human-occupied environments. The potential economic impact of robots engaged in services and manufacturing alongside humans are significant due to increased productivity and reduced costs. Another emerging area is rehabilitation and assistive robotics. The developed data-driven musculoskeletal models will also be applicable to quantification of physical impairments and estimation of muscular stress in healthcare and ergonomics. This interdisciplinary research provides excellent opportunities for undergraduate and graduate students to be engaged in analytical challenges, laboratory demonstrations of theoretical results, and experimental evaluations.

    Industrial robots in close contact with humans, robotic assistance in construction/other physical tasks, assistive devices for elderly/people with disabilities.