Anna Deichler

I’m a PhD candidate in Computer Science at KTH, focusing on multimodal machine learning for embodied agents. My research integratestext, audio, and motion, with a particular emphasis ongesture generation in situated interactions.

My recent work explores reinforcement learning andgenerative models for gesture generation, particularly inreferential communication.

In my most recent work, I introduced a framework for grounded gesture generation, combining a synthetic dataset of spatially grounded gestures and the MM-Convdataset of VR-based referential dialogues.

Looking ahead, I’m interested in combining these approaches, exploring modular architectures with pre-trained models, and leveraging language modeling for training interactive agents. My goal is to enable agents to communicate naturally and contextually using both verbal and non-verbal behavior in physical or virtual environments.