GDPO-Listener: Expressive Interactive Head Generation via Auto-Regressive Flow Matching and Group Reward-Decoupled Policy Optimization

Generating realistic 3D head motion for dyadic interactions is a significant challenge in virtual human synthesis. While recent methods achieve impressive results with speaking heads, they frequently suffer from the “regression-to-the-mean” problem in listener motions, collapsing into static faces, and lack the parameter space for complex nonverbal motions.

We propose GDPO-Listener, a framework for highly expressive speaking and listening motion generation. First, an Auto-Regressive Flow Matching architecture enables stable supervised learning. Second, Group Reward-Decoupled Policy Optimization (GDPO) isolates reward normalization across distinct FLAME parameter groups, explicitly incentivizing high-variance expressive generations.

The framework also enables explicit semantic text control for customizable responses. Extensive evaluations on the Seamless Interaction and DualTalk datasets demonstrate improved long-term kinematic variance, visual expressivity, and semantic controllability over existing baselines.