BabyBench: A Multimodal Benchmark of Infant Behaviors for Developmental AI

Francisco M. López¹, Valentin Marcel², Xavier Hinaut³, Jochen Triesch¹, Matej Hoffmann²

¹Frankfurt Institute for Advanced Studies; ²Czech Technical University in Prague; ³INRIA Bordeaux

Join the 2025 BabyBench Competition at the
IEEE ICDL Conference in Prague, Czechia

BabyBench

BabyBench is a developmental benchmark of behavioral milestones leveraging MIMo, the multimodal infant model, with a unique modular approach: environments combine behaviors, scenes, sensory modalities, actuation models, and more.

30+

Behaviors

20+

Scenes

40+

DOF

0-24

Months embodiment

4

Sensory modalities

3

Actuation models

1000+

Unique learning environments

BibTeX

        @misc{lopez2025babybench,
            title={BabyBench: A Multimodal Benchmark of Infant Behaviors for Developmental AI},
            author={Francisco M. López and Valentin Marcel and Xavier Hinaut and Jochen Triesch and Matej Hoffmann},
            year={2025},
        }