This article introduces a hybrid approach for learning Human-Robot Interaction (HRI) from Human-Human Interactions (HHI). The method uses Hidden Markov Models (HMMs) as the latent space priors for a Variational Autoencoder to model a joint distribution over the interacting agents. The interaction dynamics learned from HHI are used to learn HRI and incorporate the conditional generation of robot motions from human observations into the training. This results in more accurate robot trajectories. The generated robot motions are further adapted with Inverse Kinematics to ensure the desired physical proximity with a human. The method is verified through a user study and found to be perceived as more human-like, timely, and accurate.
Publication date: 29 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.16380
