Human Robot Handover

---

As part of a NSF funded project for the Movement Neuroscience Lab, we have been trying to understand how humans handover objects between one another, and how to recreate this complex behavior between a human and a robot. Over the 2021 Summer, our team collected a wealthy dataset of human-to-human handover in a motion capture lab. With this data, our team was able to develop a novel model for human handover based on the idea of Gaussian submovements. This model was published in a paper (which I am a co-authored) in PETRA’22: "Leveraging Submovements for Prediction and Trajectory Planning for Human-Robot Handover"

With this novel model for human handover, I integrated a motion capture system with an industrial 7-degree-of-freedom robotic manipulator for test and evaluation. This required implementing a custom velocity-based inverse-kinematics solver. The different components communicate via ROS, with a combination of data being sent over ROS’ TCP network layer, and using UDP multicast for the motion capture system. With my design, our model developers can create their models in MATLAB, and seemlessly command the robotic manipulator with a simple API. The system is currently undergoing improvements with the development of a PyQT-based user interface to control experiements to allow for future data collection.