六角未来 HexaFuture
Benchmarks

Measuring agents that interact with the physical world.

Benchmarks from HexaFuture that hand a whole real task to a frontier agent with a robot body, and score what it gets done.

All benchmarks

1 published
  • Six EmbodiedLab benches, each worked by a robot arm: air track, coupled oscillators, spring, incline, pendulums and a titration rig.
    Embodied science

    EmbodiedLabBenchmarking embodied agents that act to know

    Can an embodied agent find out how the physical world is by acting on it? 30 quantitative experiments in mechanics, electricity, optics, heat and chemistry, worked by a robot arm in a simulated lab.

    30
    Experiments
    14
    Apparatus
    5
    Disciplines
    6
    Frontier models
    View benchmark PaperCode