Leveraging Foundation Models for Robust Domestic Autonomy

Speaker: Eric Martinson

Domestic environments present unique challenges for service robotics: they are rich in visual and task complexity, but often require minimalistic solutions to reduce cost and avoid interfering with the people who live there. This talk examines the role of robots in the home and highlights current efforts at LTU to use foundation models as a bridge toward the next generation of capable, reliable domestic autonomy.

Ongoing research in Computer Science and Robotics at LTU focuses on:

• Advances in open-vocabulary object detection, exploring how foundation models can expand recognition, improve generalization, and support more adaptive perception in cluttered, dynamic homes.
• Creating safe, high-fidelity simulation spaces derived from individual homes, enabling robots to explore, learn, and refine behaviors before deployment.
• Using large language models to interpret user intent and guide robotic task selection, improving both the robustness and naturalness of domestic robot behavior.

Together, we are working toward a future where domestic robots understand the world more flexibly and can provide meaningful, trustworthy support in everyday home environments.