artifactarena2026MIT

Artifact Arena

A benchmark dataset of robots designed by frontier AI models for MuJoCo sumo competition, containing 170 bot designs with their MuJoCo bodies, controllers, and tournament game results across 123,832 total matches.

Downloads29
Episodes123832

Why This Matters for Physical AI

Artifact Arena evaluates how well frontier AI models can design functional robots for physical tasks, providing a benchmark for assessing embodied AI capabilities through competitive morphology generation and control.

Technical Profile

Modalities
simulation
Robot Embodiments
custom_morphologies
Environment
simulation
Task Types
competitionsumo
Episodes
123832
Data Format
CSV
Annotation Types
game_resultselo_ratingsdesign_texts
License
MIT
Part of the Artifact Arena family

Access

Need custom simulation data?

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

Request a Sample Pack

Related Datasets