This paper presents a novel learning-based approach to dynamic robot-to-human handover, addressing the challenges of delivering objects to a moving receiver. We hypothesize that dynamic handover, where the robot adjusts to the receiver’s movements, results in more efficient and comfortable interaction compared to static handover, where the receiver is assumed to be stationary. To validate this, we developed a nonparametric method for generating continuous handover motion, conditioned on the receiver's movements, and trained the model using a dataset of 1,000 human-to-human handover demonstrations. We integrated preference learning for improved handover effectiveness and applied impedance control to ensure user safety and adaptiveness. The approach was evaluated in both simulation and real-world settings, with user studies demonstrating that dynamic handover significantly reduces handover time and improves user comfort compared to static methods. Videos and demonstrations of our approach are available at https://zerotohero7886.github.io/dyn-r2h-handover/.
Baseline
Baseline
(Ours)
"We introduce a novel framework for dynamic robot-to-human handover that mimics human-to-human interactions."
Many previous works assumed that the receiver is stationary, but in real-world scenarios, the receiver is often moving. We hypothesize that dynamic handover, where the robot adjusts to the receiver's movements, results in more efficient and comfortable interaction compared to static handover, where the receiver is assumed to be stationary. To implement the seamless and natural interaction, we leverage human-to-human handover demonstration data and preference learning to train the robot.
To generate natural robot motions, we take inspiration from human interactions. We collect 1,000 scenarios of human-to-human handover data using a combination of motion capture and a human pose estimation system. The collected human motion data enables the robot to learn more intuitive handover movements for humans.
Failure case: "Too fast"
Failure case: "Too slow"
Failure case: "Too strong"
Our system employs force-based control, where key parameters related to tracking performance, stiffness, and release timing are intricately interdependent. To optimize the system, we utilize a preference learning approach that adapts these parameters based on human feedback.
"No look pass"
"Forgot my cellphone"
"Long time no see"
We conducted a number of user studies to evaluate the effectiveness of our approach. The results show that dynamic handover significantly reduces handover time and enhances user comfort compared to static methods and a dynamic approach without system optimization.