Models & learning
SMPL-X
SMPL-X is a learned parametric 3D human model that represents a unified articulated body, hands, and expressive face using shape and pose parameters. It provides a consistent mesh and skeleton, but fitted SMPL-X motion is an estimate derived from observations and model assumptions rather than direct measurement of every joint or surface point.
Also known as: SMPL-X body model, SMPL-X model
Updated
One model connects body, hands, and face
The SMPL-X paper extends parametric human modelling into a unified mesh with articulated body, hands, and face. Shape parameters change the person-specific geometry, while pose parameters articulate the kinematic structure and drive pose-dependent deformation.
That shared representation is useful when a dataset needs whole-body posture and detailed hands in one coordinate system. It also supports consistent surface points for visualisation, contact analysis, self-intersection tests, and transfer between recordings.
Parameters are not raw measurements
An SMPL-X sequence is normally fitted or inferred from cameras, trackers, gloves, or another body representation. Its quality therefore depends on calibration, sensor visibility, the fitting objective, motion prior, body-shape estimate, and coordinate-frame conventions.
A plausible mesh can still place an elbow incorrectly, penetrate an object, slide a foot, or smooth away a fast contact. Dataset documentation should distinguish directly measured inputs from fitted SMPL-X parameters and state whether body shape is fixed per participant, per session, or per frame.
File conventions need to be explicit
Pose may be stored as local axis-angle rotations, rotation matrices, quaternions, or another representation. Global translation may refer to the pelvis, model origin, or an additive library parameter. Scale, world orientation, handedness, up axis, joint order, and the hand-pose convention also affect interpretation.
EgoExplorer stores a metric world pelvis position, world-relative root orientation, local body-joint rotations, and local finger rotations while referencing participant shape and mount calibration by profile ID. Publishing those conventions prevents a numerically valid array from being replayed in the wrong place or frame.
Human motion still needs robot retargeting
SMPL-X describes a human body, not a robot. A humanoid may have different link lengths, joint axes, limits, feet, hands, and actuation. Motion transfer therefore needs a separate retargeting step plus collision, balance, contact, and controller validation.
A dataset should retain both the original human parameters and any derived robot reference. Keeping them separate preserves the mapping error and avoids presenting a retargeted trajectory as if it were directly measured robot motion.
Sources
Related terms
Models & learning
Pose estimation
Pose estimation is the process of inferring the position and orientation of a body, object, camera, hand, or robot relative to a specified coordinate frame. In three-dimensional robotics this is often called 6D or 6-DoF pose estimation because the result has three translational and three rotational degrees of freedom, even when orientation is stored with more than three numbers.
Data & collection
Motion capture
Motion capture is the measurement and reconstruction of a person’s or object’s movement over time, commonly as joint positions, orientations or a fitted body model. Optical markers, cameras and inertial sensors can supply the measurements. Humanoid robotics uses the resulting motion sequences for analysis, imitation and retargeting to a robot body.
Data & collection
Human–object interaction
Human–object interaction is the physical and semantic relationship between a person and an object while the person observes, reaches, grasps, moves, uses, or otherwise acts on it. In robotics datasets, the term often refers to recordings and annotations that connect human body or hand motion with object identity, pose, contact, action, and task context.
Simulation & transfer
Motion retargeting
Motion retargeting is the adaptation of a recorded or generated motion from one body to another with different proportions, joints or limits. For humanoid robots, it maps source poses or trajectories into robot configurations while preserving task-relevant relationships such as contacts and end-effector paths and satisfying kinematic, balance, collision and actuator constraints.
Data & collection
Sparse motion capture
Sparse motion capture reconstructs an articulated body's pose from a small set of measurements, such as head, wrist, and ankle tracker poses, instead of directly observing every joint. Unmeasured joints are inferred with kinematic constraints or learned motion priors, so the result is a model-based estimate rather than direct sensor ground truth.