Models & learning
Motion tracking
Motion tracking is a control task in which a robot policy follows a supplied reference pose or trajectory over time while maintaining physical stability. Success measures how closely and reliably the robot reproduces that reference; it does not by itself show that the robot can perceive a goal, choose a task, or plan a motion.
Also known as: reference motion tracking, motion imitation tracking
Updated
Reference in, control action out
A motion tracker receives a reference that describes how a body should move. The reference may contain target joint configurations, body-keypoint positions, root motion, or a combination of those signals. The policy compares that target with the robot’s current state and produces actions that reduce the tracking error without causing a fall or violating control constraints.
DeepMimic established a widely used reinforcement-learning formulation in which a simulated character learns to imitate reference motion clips while satisfying physical dynamics. Humanoid trackers use the same broad idea with robot-specific observations, actions, rewards, and transfer requirements.
Tracking, retargeting, and planning are different
Motion retargeting converts movement from a source body into a reference for a target robot. Motion tracking is the next problem: making the physical or simulated robot follow that reference. A trajectory can be well retargeted yet remain difficult to track because of contact, actuator, latency, or model errors.
Task planning sits above both. A planner or higher-level policy decides what outcome to pursue and which motion to request. A tracker can reproduce a digging movement when given the reference without knowing where to dig, what material is present, or whether the task succeeded.
What a tracking result should report
A useful evaluation names the robot, control frequency, reference source, held-out conditions, failure rule, and trial count. Common metrics include stable completion rate, joint-position or velocity error, body-keypoint error, falls, and external pose error.
“Zero-shot” should identify what was unseen. In the Humanoid-GPT study, the held-out object is the target motion, while the embodiment remains a Unitree G1. That is evidence of motion generalization, not automatic transfer to a robot with different joints, mass, actuators, or control interfaces.
Sources
Related terms
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
Trajectory
A trajectory is a time-ordered sequence of states or observations, actions and, where applicable, rewards generated as an agent or robot evolves. A complete episode or policy rollout often yields a trajectory, but the terms are not universally identical: trajectories may be partial, while episodes have dataset- or environment-defined boundaries.
Hardware & control
Whole-body control
Whole-body control coordinates a humanoid’s full multibody state, actuated joints and environmental contacts to pursue several motion or force objectives while respecting constraints such as balance, joint limits and friction. It is a family of hierarchical or optimisation-based methods, not one algorithm; implementations may output joint positions, accelerations or torques.
Hardware & control
Feedback control
Feedback control is a closed-loop control method that measures a system’s current output or state, compares it with a target and adjusts the command using the resulting error. In robotics, feedback can correct joint, end-effector, balance or force errors as new sensor measurements arrive.