Diffusion-Based Imitation Learning for Social Pose Generation

TL;DR

This work introduces a diffusion-based imitation learning approach to generate facilitator nonverbal social cues from multi-person pose data, adapting diffusion-based behavior cloning with a transformer-based denoiser and the Diffusion-X sampling strategy ($T=50$, $M=8$). It compares two conditioning representations—raw image observations and pre-processed pose keypoints plotted on a white background—using a dataset of multiparty social interactions (FUMI-MPF) captured with a 360° camera, totaling ~90k frames. Mean per joint position error (MPJPE) and per-frame processing time are used to assess accuracy and efficiency, revealing that pose-keypoint conditioning improves performance at the cost of higher computation, while real-time feasibility remains a challenge. The results suggest diffusion-based BC is a viable path for generating contextually relevant social poses, with future work focused on scaling to larger groups, optimizing feature extraction, and reducing latency for real-time deployment.

Abstract

Intelligent agents, such as robots and virtual agents, must understand the dynamics of complex social interactions to interact with humans. Effectively representing social dynamics is challenging because we require multi-modal, synchronized observations to understand a scene. We explore how using a single modality, the pose behavior, of multiple individuals in a social interaction can be used to generate nonverbal social cues for the facilitator of that interaction. The facilitator acts to make a social interaction proceed smoothly and is an essential role for intelligent agents to replicate in human-robot interactions. In this paper, we adapt an existing diffusion behavior cloning model to learn and replicate facilitator behaviors. Furthermore, we evaluate two representations of pose observations from a scene, one representation has pre-processing applied and one does not. The purpose of this paper is to introduce a new use for diffusion behavior cloning for pose generation in social interactions. The second is to understand the relationship between performance and computational load for generating social pose behavior using two different techniques for collecting scene observations. As such, we are essentially testing the effectiveness of two different types of conditioning for a diffusion model. We then evaluate the resulting generated behavior from each technique using quantitative measures such as mean per-joint position error (MPJPE), training time, and inference time. Additionally, we plot training and inference time against MPJPE to examine the trade-offs between efficiency and performance. Our results suggest that the further pre-processed data can successfully condition diffusion models to generate realistic social behavior, with reasonable trade-offs in accuracy and processing time.

Figures (3)

• Figure 1: Diffusion BC model flow chart
• Figure 2: Two types of image used to test model performance
• Figure 3: Processing Time vs. MPJPE for Raw Images and Plotted Images