Video to humanoid motion data

Video-to-humanoid facts

• The output should be reviewed before any downstream robot adaptation.
• AIMoCap separates robot-oriented targets from Default animation output.
• Clear source video and trim windows are especially important for robot motion review.
• Humanoid feasibility depends on robot morphology, contact handling, timing, and balance constraints.
• This workflow is useful for exploration and iteration, not direct hardware execution.
• Video-to-humanoid-motion output is not a direct robot-control or hardware-command stream.
• A humanoid-motion review should compare human intent, target artifact, contact timing, center-of-mass behavior, and simulation outcome before any hardware discussion.
• If the target robot has different limb proportions or joint limits, downstream retargeting may need to change the motion even when the source solve looks plausible.
• A source clip can pass visual motion review but still fail humanoid robot review because the robot cannot match the same reach, stance, speed, or contact pattern.
• Teams should label each attempt by outcome: usable reference, needs retargeting edits, source video not readable enough, or rejected by robot constraints.
• A video-to-humanoid-motion handoff should name the target morphology, validation environment, and failure category before the result is reused as training or test data.
• If the downstream target is not Unitree G1, treat the page as integration planning and avoid assuming the same artifact semantics transfer to a different humanoid.
• A humanoid validation packet should separate source-video evidence, generated motion artifact, simulator/controller settings, target morphology notes, and the final accepted or rejected decision.
• Unitree G1-oriented validation is useful as a concrete public target, but it should not be treated as automatically transferable to every humanoid with different proportions, feet, joints, or controller behavior.
• Video-to-humanoid review should label the adaptation layer separately from the generated artifact, because downstream changes may fix timing, stance, contact, joint limits, or controller tracking.
• A clip should not be reused as humanoid training or test data unless the record says which target morphology and validation environment accepted it.
• If downstream edits are required, the record should preserve both the original artifact and the adapted version so reviewers can see what changed after AIMoCap.