Pose Generation for Social Robots in Conversational Group Formations

Marynel Vázquez, Sasha Lew, Eden Gorevoy and Joe Connolly

Yale University

We study two approaches for predicting an appropriate pose for a robot to take part in group formations typical of social human conversations subject to the physical layout of the surrounding environment. One method is model-based and explicitly encodes key geometric aspects of conversational formations. The other method is data-driven. It implicitly models key properties of spatial arrangements using graph neural networks and an adversarial training regime. We evaluate the proposed approaches through quantitative metrics designed for this problem domain, and via a human experiment. Our results suggest that the proposed methods are effective at reasoning about the environment layout and conversational group formations. They can also be used repeatedly to simulate conversational spatial arrangements despite being designed to output a single pose at a time. However, the methods showed different strengths. For example, the geometric approach was more successful at avoiding poses generated in non-free areas of the environment, but the data-driven method was better at capturing the variability of conversational spatial formations. We discuss ways to address open challenges for the pose generation problem and other interesting avenues for future work.

Instructions

This website illustrates the two pose generation methods, geometric and data-driven. Several maps derived from the iGibson simulation environment are provided. These environments contain simulated social contexts with one to six existing individuals.

To generate a pose, you can select an environment map and the number of generated poses. The social context is chosen randomly from the simulated dataset on the selected environment map. Single pose predictions are plotted directly. When a distribution of poses is predicted, we search for a mode of the predicted locations using mean shift, and output the pose in the distribution that is closest to the mode.

Parameter Selection

Environment Number of Poses

Results

Unknown space

Tall obstacle

Short obstacle

Free space

Predicted pose

Prediction distribution

Context

Geometric Method

Data-driven Method