Fall 2026 GRASP Seminar: Research Overview Day 2
CJ Taylor, professor of Computer and Information Science and GRASP director at the University of Pennsylvania, opens a research overview with presentations from GRASP faculty and students on robotics, AI, and rehabilitation.
In this talk
- 0:01
Perception, learning, and health AI
CJ Taylor describes mapping large outdoor areas with depth panoramas and navigation for agricultural robots working in vineyards and orchards, including GPS-denied settings. Darian presents René Vidal’s group’s research in machine learning, trustworthy AI, computer vision, and health, including blood-cell analysis and assessing children’s ability to imitate actions.
- 12:04
Robots that manipulate and support rehabilitation
Rachel Holiday’s Autonomous Manipulation Lab studies long-horizon tasks such as cleaning, cooking, and handling objects under uncertainty, using interaction to gather information and adapt plans. Eric Waters presents Michelle Johnson’s Rehabilitation Robotics Lab projects, including infant assessment for cerebral palsy, robot-assisted therapy, haptic learning between people, and EEG and EMG studies of post-stroke arm use.
- 22:32
Field robotics, robot learning, and contact
Rand Clara outlines Vijay Kumar Lab work on UAV autonomy, precision agriculture, event cameras, air-ground collaboration, and orchard sensing that tracks fruit growth. U.S. Joe describes research on learning manipulation from few demonstrations and explains how modeling action variability can improve on L2 regression for robot policies. Michael Posa discusses using contact information to infer object properties and model-based control that handled multiple objects without retraining.
- 41:33
Soft, modular, and perception-guided robots
Cynthia Sun’s group studies compliant robots, from coordinated underwater swimmers to soft conduits that can pump blood, and designs robots through folding and printed flexible circuits. Mark Yim presents projects including the modular CHAMPS snake robot, a tiny drone, reconfigurable robots, steerable needles, and twisted-string actuators. Another lab overview covers OmniGuide for spatially precise robot policies, Mimic for learning dexterous motion from human video, active perception, and event cameras.
- 59:37
Creative AI and physical intelligence
Kate presents VibeSpace, a method for finding curved visual paths between images while preserving attributes such as hairstyle, shape, or expression; the group is also studying how people develop humor captions. Jant’s group aims to build physical intelligence by combining video and 3D world models with latent-space reasoning and robot action, supported by research on generative models.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.