velocityaccelerationtrajectoryinteractionintenttemporal context
AI that understands how things move.
Kinetiq is building a foundation model for motion — learning the dynamics of humans, animals, machines, vehicles, and robots from billions of trajectories.
Talk to KinetiqOne model. Every motion.
velocityaccelerationtrajectoryinteractionintenttemporal context
Vision models learned the world in frames. We are learning it in motion.
Motion is not a sequence of images. It is a structured signal with its own language.
IMAGEspatial only
appearancegeometryposition
MOTIONspatial + temporal
directionvelocityaccelerationinteractiontimingintent
Different bodies. Different machines. The same underlying language of motion.
A human step, a quadruped gait and a legged robot's stride look nothing alike. Under a shared representation, their temporal structure converges.
fig. 02 — trajectory familiesscroll to align families
From raw trajectories to a universal motion representation.
fig. 01 — system overviewkinetiq / motion encoder
VIDEO
IMU
MOTION CAPTURE
ROBOT TELEMETRY
WEARABLES
VEHICLE SENSORS
INDUSTRIAL SENSORS
raw trajectories
KINETIQ MOTION ENCODER
temporal attentioncross-body alignmentphysics-aware objectives
SPATIOTEMPORAL REPRESENTATION
MOTION GENOME
downstream tasks
PREDICTION
CLASSIFICATION
CONTROL
SIMULATION
ANOMALY DETECTION
FORECASTING
The same model can read very different worlds.
Hover a domain to re-condition the motion field.
Understanding motion means predicting what happens next.
fig. 03 — forward rolloutillustrative
temporal context
2.4s
prediction horizon
1.8s
confidence field
dynamic
Movement becomes searchable.
motion → representation → similarity
fig. 04 — embedding spaceprojection: 2d conceptual
The smallest deviation can contain the biggest signal.
Once a population of motion shares a representation, divergence becomes measurable — across bodies, machines and processes alike.
Motion is everywhere. The training data is fragmented.
A general motion model requires learning across domains, bodies, environments, and scales.
One motion model. Many physical applications.
KINETIQ
MOTION GENOME
Motion EmbeddingsTrajectory PredictionMotion ClassificationAnomaly DetectionBehavior ModelingTemporal ForecastingMotion GenerationMOTION GENOME
ROBOTS
PEOPLE
VEHICLES
MACHINES
LEARNREPRESENTPREDICTGENERALIZE
The physical world is a stream of motion.
KINETIQ
Motion Genome
A foundation model for the language of movement.