Data & collection
Cross-embodiment data
Cross-embodiment data is robot training data drawn from multiple physical embodiments, such as arms, mobile manipulators, quadrupeds or humanoids with different kinematics, sensors and action spaces. The datasets are aligned or packaged so models can learn jointly from experience produced by different robots, although shared formatting does not make their observations or controls physically equivalent.
Also known as: X-embodiment data, multi-embodiment robot data
Updated
What crosses between embodiments
The phrase describes the composition of the data, not a single sensor format or learning method. An embodiment may differ in link geometry, degrees of freedom, gripper or hand design, camera placement, proprioceptive signals and control interface. The Open X-Embodiment project pooled datasets from many robots into a consistent episode format, while retaining differences in their observations and action spaces.
Making those records trainable together requires explicit alignment decisions. A pipeline might choose a common end-effector action representation, normalise actions per dataset, select comparable camera views or mask unavailable modalities. These transformations improve interoperability, but they do not erase the physical differences between robots.
Why the data is useful
Cross-embodiment mixtures let a model learn from tasks, objects and environments that no single robot dataset covers. The RT-X experiments reported positive transfer from co-training across robots, while Octo used a multi-robot mixture and was designed for fine-tuning to new observations and action spaces.
That evidence does not mean trajectories from one platform can be replayed directly on another. A target humanoid may still need action adapters, retargeting, in-domain demonstrations or policy fine-tuning.
What a usable record should identify
Each episode should preserve the source robot, kinematic model, sensors, coordinate frames, control mode, action semantics, rates and any normalisation applied. Without that provenance, a shared file format can conceal incompatible commands and make transfer results difficult to interpret.
Sources
Related terms
Models & learning
Robot foundation model
A robot foundation model is a broadly pretrained model intended to provide a reusable starting point for multiple robot tasks, environments or embodiments. It learns from diverse robotics and sometimes web or human data, then acts directly or is adapted with target-domain data. The term describes a training and reuse strategy, not one fixed architecture.
Models & learning
Vision-language-action model
A vision-language-action model is a robot policy that conditions on visual observations and natural-language instructions to predict executable actions. It adapts a pretrained vision-language model, or a similar multimodal backbone, using robot trajectories so that semantic knowledge is connected to an embodiment’s action space. The action representation may use discrete tokens or continuous values.
Data & collection
Demonstration
A demonstration is a recorded example of how an intended task or behaviour is performed, usually represented as a time-aligned sequence of observations, states and actions. For humanoid robot learning, demonstrations may come from teleoperation, kinaesthetic guidance, motion capture or autonomous experts and provide targets for imitation.
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.