Evidence before adjectives
Scale, rights, modalities, coverage, and quality claims resolve to records.
Company thesis
The data layer for models that act
ActionTrajectories turns cause and effect into inspectable collection, release, and licensing contracts for world models.
01
00.000sObserve
rgb.player_view02
00.050sAct
move.forward + camera.delta03
00.100sTransition
pose.x +0.84 m04
00.150sOutcome
progress 0.31 → 0.36The values are synthetic and illustrate the public specification. They are not measurements from a commercial dataset.
Why this exists
The internet gave AI language, images, and passive video. It did not reliably record which action caused the next state. That gap becomes critical when a model must predict, plan, or act inside a changing world.
Our starting wedge is instrumented gameplay and simulation: controlled environments with dense behavior, explicit inputs, measurable outcomes, and privileged state. The same trajectory contract extends to robotics, vehicles, and other embodied systems.
ActionTrajectories focuses on the sequence that matters: observation, action, next state, and outcome, with timing, provenance, quality, and rights attached.
Missing stays missing
Unknown state, inferred action, and unresolved rights are labeled as such.
Open is not inventory
External research references never masquerade as first-party commercial data.
The buyer gets an interface
Schema, loaders, tests, provenance, known limits, and release records ship together.
Legal operator
10xme Technology Inc.
ActionTrajectories is an independent brand operated by 10xme Technology Inc. Public reference datasets remain owned, hosted, and governed by their respective creators and providers.
Review site termsBring us the training target.
We will map it to the observations, actions, state, outcomes, rights, and delivery evidence the dataset must contain.