We are a research group at the MLT-Lab at DFKI specialized in dialogue processing for human-robot collaboration and for teamwork support.
The goal of dialogue processing is to build systems that can interpret and participate in natural spoken dialogue interaction. Dialogue processing comprises automatic speech recognition, natural language understanding and interpretation of multimodal inputs, dialogue state modeling, dialogue management, generation and synthesis of verbal and multimodal outputs. Dialogue is situated, social and personal. We therefore also model how what is said relates to and is grounded in the context of preceding interaction(s), knowledge about the world and the physical situation; what roles the dialogue participants have and what relationships exit between them; and how the current interaction reflects accumulated interpersonal experience.
We have extensive experience with dialogue processing in a variety of applications, including: home and office service; healthcare; human-robot collaboration in hybrid industrial production teams; robot-assisted disaster response. We have integrated dialogue processing into mobile robot-based systems in various EU and national projects. In these contexts we have worked intensively with end users, including first responders (e.g., firefighters), patients (e.g., diabetic children) and caregivers.
A-DRZ is setting up a competence center bundling scientific and technical expertise in robot-assisted disaster response. The TR group develops methods for interpreting verbal communication between the members of the response team and anchoring it in mission process models.
Project website: rettungsrobotik.de
NotAs investigates the use of language technology to support emergency call operators in handling foreign language calls. The TR group develops machine translation and interpretation of emergency calls.
IMPRESS (BMBF; 2020-2023) investigates the integration of semantic knowledge into linguistic and multimodal embeddings and the impact on selected downstream tasks in the processing of speech and image/video.
INTUITIV investigates intuitive nonverbal and informative verbal robot-human communication. The TR group develops methods for situated spoken natural language interaction with focus on indoor navigation dialogue.
PAL (EU; 2015-2019) developed an integrated system providing comprehensive, prolonged, personalized and context-sensitive support in order to advance the self-management of children with Type 1 diabetes, so that adequate shared patient-caregiver responsibility is established before the child reaches adolescence. The TR group contributed dialogue management, long-term interaction memory and spoken input and output processing.
Concept overview video: https://www.youtube.com/watch?v=v_7RqBq6tv0
Youtube channel: https://www.youtube.com/channel/UC7gsnMoFvw9Nztvk0XmSSBQ
TRADR (EU; 2013-2017) investigated how a team consisting of humans and several ground and airborne robots develops situation awareness and experience gradually over multiple synchronous and asynchronous sorties in a long-term mission. TRADR addressed persistence of environment models, multi-robot action models and human-robot collaboration over time. The TR group contributed dialogue processing for team communication and interactive mission reporting.
TRADR Joint Exercise Dortmund 2015: mission concept: https://www.youtube.com/watch?v=4z86nUlgEqc
TRADR Joint Exercise 2016 Prague: overview of technology: https://www.youtube.com/watch?v=mJpLdiC75ns
Final review: https://www.youtube.com/watch?v=W8onMryGTEA
View more on TRADR youtube channel: https://www.youtube.com/channel/UCvrPijjrCFGWdJIHkKbFK-A
HySocieTea (BMBF, 2014-2016) examined collaboration and communication in teams consisting of technologically augmented humans, autonomous robots, virtual characters and softbots, working on joint tasks in flexible production processes. The TR group contributed to the development of an integrated team collaboration system, including the processing for spoken dialogue between humans and artificial agents and otology-based modeling and reasoning for contextual reference resolution and generation.