Spring 2021 GRASP SFI: Monroe Kennedy III - March 17th, 2021

In this talk

1. 5:34

   What makes a strong team

   Kennedy argues that effective teammates need shared objectives, ways to anticipate one another, and a clear sense of each member’s role. Human teams coordinate through speech, sight, and touch, each with different strengths and limits. In a lane-change example, trust depends on both drivers wanting to avoid a crash and knowing the other has the ability to respond.

2. 10:26

   Sensing and modeling contact

   Kennedy reviews robotic control, perception, and decision-making, with particular attention to force control and tactile sensing. His lab is developing a curved fingertip sensor that uses a fisheye lens and colored illumination to estimate surface deformation, normal force, and shear. The longer-term goal is to use those measurements to plan stable grasps and in-hand manipulation while accounting for uncertainty.

3. 22:03

   Wearables that restore or extend ability

   Kennedy describes an intelligent prosthetic that could combine task context, gaze, and limited muscle signals to infer what a user wants to do. He also outlines a wearable fall-warning sensor that uses visual-inertial motion history and environmental observations to predict risky paths, plus a mobile sign-language translation project that accounts for expression and gesture sequences.

4. 38:08

   Robots as physical teammates

   For cooperative transport, Kennedy’s group studies how people convey intent through forces on a shared object and is building from an online multi-person simulator toward physical robot collaboration. In box-carrying tasks, a robot could observe who carries which loads and take heavier boxes if that improves the team’s pace. He also discusses learning task structure to reduce the data needed to teach a robot complex work.

5. 41:28

   Safe handoffs and practical lessons

   Kennedy considers how a self-driving vehicle can transfer control to a person safely, using observations of the driver’s behavior to estimate whether they can remain within a safe operating region. He concludes that robots should model both human behavior and task roles, and that better audio, visual, and tactile communication could help them augment or restore human capabilities. In the discussion, he notes that partners may start from a general model and adapt it through ongoing interaction.