Read the Room
A public challenge: watch a short scene, stop when you can read the highlighted person’s next move, then make your guess.
How early can you tell what someone will do next?
Humans read movement almost without thinking - it's what we know as 'human instinct'. Robots don't come pre-installed with that instinct.
Intent Horizon is being built for safer human-robot interactions by benchmarking the moment a future action becomes clear.
Why Intent Horizon
Earlier understanding can lead to safer interaction.
Robots and autonomous systems often need to respond before another person’s movement is complete. The challenge is recognising intent early enough to act safely, while uncertainty still exists.
Intent Horizon will provide shared scenarios, definitions and scoring so people, models and research teams can be compared against the same evidence and the same rules.
Here’s what’s coming
A benchmark designed to be understood, tested and improved.
Research participation
A longer, more rigorous study using a broader and carefully balanced set of scenarios under a consistent protocol.
An open dataset
Versioned scenarios, target tracks, motion representations, candidate intents, verified outcomes and reproducible methodology.
A transparent leaderboard
Verified human and model results tied to the exact benchmark release, evidence window and scoring policy used.
The benchmark standard
Fairly and squarely.
Same evidenceComparable observation windows and intent choices.
Early and reliableRewarding correct predictions that remain stable—not lucky spikes.
Open methodPublished definitions, releases, exclusions and evaluation rules.