Anonymous-nips-submissions2026cc-by-4.0

MiraBench

An evaluation dataset for assessing action-conditioned reliability in robotic world models, containing tests for action following fidelity, optimism bias detection, physical consistency, and physics law compliance.

Downloads30

Why This Matters for Physical AI

MiraBench provides systematic evaluation benchmarks for validating the reliability and physical correctness of learned world models, which is critical for safe deployment of embodied AI systems.

Technical Profile

Modalities
video
Action Space
action-conditioned
Task Types
world_model_evaluation
Annotation Types
action_labelsvalidation_labels
License
cc-by-4.0
Part of the MiraBench family

Access

Need custom video data?

Claru builds purpose-built datasets for any environment applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets