Cross-embodiment

Extracted Attributes

• Field

  Robotics and Artificial Intelligence

• Key Advantage

  Overcomes the sim-to-real gap and reduces per-platform training costs

• Core Techniques

  Teleoperation, Motion-invariant losses, World Models, Canonicalization, Multi-robot datasets

• Primary Objective

  Generalize control policies and skill transfer across heterogeneous robot morphologies and sensory-action spaces

Timeline

• Mirage zero-shot cross-embodiment policy transfer method presented at Robotics: Science and Systems conference (Source: Web Search)

  2024-07-15

• Cross-embodiment learning methodologies highlight standardized spaces and 3D world models for heterogeneous robot platforms (Source: Web Search)

  2026-01-22

• 1X plans launch of NEO humanoid robot utilizing cross-embodiment and world models for foundation AI integration (Source: Document c4c99c45-ab82-495f-b76a-fc9ae670bc0d)

  2026-01-01

Cross-dressing

Cross-dressing means wearing clothes traditionally or stereotypically associated with a different gender, such as men wearing women’s clothes, or more rarely women wearing men’s clothes. It is a form of dress or presentation rather than any particular gender identity, sexual orientation, or motivation, and has appeared in many societies across history for reasons including disguise, comfort, self-expression, ritual, and performance. Cross-dressing is not the same as being transgender, though older writings often used broader and less precise terminology that grouped different experiences together. The term and its related vocabulary have changed over time. In modern usage, cross-dressing is generally preferred to older terms such as transvestism or transvestite, which were historically used in medical and psychiatric contexts and are now often regarded as outdated or offensive. It has also been regulated or criminalised in various places, with laws against cross-dressing historically used to enforce gender norms and to target transgender and gender-nonconforming people. Cross-dressing has played a notable role in theatre, religion, entertainment, and popular culture, including traditions such as kabuki, drag, and British pantomime. Socialization establishes social norms among the people of a particular society. With regard to the social aspects of clothing, such standards may reflect guidelines relating to the style, color, or type of clothing that individuals are expected to wear. Such expectations may be delineated according to gender roles. Cross-dressing involves dressing contrary to the prevailing standards (or in some cases, laws) for a person of their gender in their own society. For thousands of years, people have cross-dressed in order to disguise themselves or be recognised as another gender, for reasons varying from comfort, entertainment, or self-expression. Cross-dressing has appeared in theater performance traditions globally, including kabuki and drag, as well as British pantomime, where it is traditionally utilized for comedic effect.

Web Search Results

• Cross-embodiment - Humanoid

  # Cross-embodiment ## Cross-embodiment Cross-embodiment refers to training or adapting robotic policies across different physical platforms. Instead of developing a control model for a single robot, engineers aim to generalize behaviors across bodies with varying limb lengths, joint types, or actuation systems. The goal is to reduce per-robot training cost and improve skill transfer. Recent models like RT-X and Open X-Embodiment use multi-robot datasets to align control across morphologies. Policies are conditioned on embodiment-specific inputs while sharing structure across tasks. In simulation, this enables curriculum learning and zero-shot transfer. In real-world use, it supports faster deployment across fleets or hardware revisions. [...] Cross-embodiment is now central to scaling humanoid intelligence: when one robot learns, the whole system improves. HUMANOID ## Building a future where robots amplify human potential Copyright © 2024-2026 SKL Robotics LTD. ### Product ### Technology ### Company ### News ### ENQUIRE ### CONNECT ### LEGAL

• Cross-Embodiment Learning in Robotics

  Cross-embodiment learning is a field in robotics and machine learning concerned with designing policies, representations, and learning systems that can transfer knowledge and skills across heterogeneous robot morphologies, sensor configurations, and action spaces. Unlike traditional approaches that require retraining for each specific physical platform, cross-embodiment methods seek true generalization across robots (or between humans and robots) that differ in kinematics, morphology, control modalities, and sensory observations. This paradigm is motivated by the need for scalable, cost-effective robot learning and the aspiration to build generalist robots capable of leveraging data and demonstrations from fundamentally different bodies. [...] 2000 character limit reached # Cross-Embodiment Learning in Robotics Updated 22 January 2026 Cross-embodiment learning is the study of transferring robotic skills across different morphologies using standardized action and observation spaces. It leverages motion-invariant losses and canonicalization techniques to neutralize embodiment-specific dynamics and sensor variations. Advanced systems combine multi-modal fusion and 3D world models to enable robust zero-shot and few-shot policy transfer across diverse platforms. [...] A foundational principle in cross-embodiment learning is the explicit construction of unified action and observation spaces. Policy architectures such as those in LEGATO unify the observation and action spaces by collecting all demonstrations—regardless of whether the demonstrator is human or robot—using a standardized, instrumented manipulation tool (e.g., the LEGATO Gripper). The gripper design, equipped with an egocentric stereo camera and actuation interface, ensures identical camera viewpoints, actuation speed, and grasp kinematics across all demonstrations and subsequent deployments. This abstraction allows a single high-level visuomotor policy π\_H to be trained on the shared data and then transferred, without further per-robot learning or calibration, to arbitrarily different

• Scaling Cross-Embodiment World Models for Dexterous Manipulation

  Cross-embodiment learning seeks to build generalist robots that operate across diverse morphologies, but differences in action spaces and kinematics hinder data sharing and policy transfer. This raises a central question: Is there any invariance that allows actions to transfer across embodiments? We conjecture that environment dynamics are embodiment-invariant, and that world models capturing these dynamics can provide a unified interface across embodiments. To learn such a unified world model, the crucial step is to design state and action representations that abstract away embodiment-specific details while preserving control relevance. To this end, we represent different embodiments (e.g., human hands and robot hands) as sets of 3D particles and define actions as particle displacements, [...] Cross-embodiment learning underpins the vision of building generalist embodied agents that operate across diverse physical embodiments by learning from heterogeneous data [3, 46]. This paradigm is becoming increasingly important as variations in morphology arise naturally from manufacturing tolerances, wear, and hardware upgrades, while large-scale robot deployments generate diverse cross-embodiment datasets (e.g., [22, 11]). Prior progress has shown embodiment-level generalization in locomotion and in manipulation with parallel grippers [4, 17, 43, 23], whereas dexterous manipulation has largely been limited to grasping [42, 10] and in-hand reorientation . Extending such generalization to broader task domains, such as non-prehensile object pushing and deformable object manipulation, [...] One key goal of cross-embodiment learning is to learn from diverse embodiments, with learning from human data being a special case. Most existing approaches are model-free, which learn mappings from observations to actions end-to-end via reinforcement learning (RL) or behavior cloning (BC). Prior work includes using human demonstrations to guide RL for dexterous manipulation [27, 6], cross-embodiment RL training for locomotion , and scaling to many embodiments through a combination of RL and BC [3, 25]. When high-quality demonstrations are available, BC can provide stable learning signals, which powered recent efforts on training vision-language-action (VLA) models on cross-embodiment datasets [4, 17, 23, 54, 49]. Human demonstrations can also be adapted to robots via motion retargeting

• Improving Compositional Generalization in Cross- ...

  Abstract Cross-Embodiment Learning (CEL) aims to train a generalist pol-icy model by integrating large-scale compositional interactions of heterogeneous agents and environments. However, the inherent conflict between the unbounded space of agent-environment com-binations and a single unified policy model hinders generalization to unseen combinations. To address this challenge, we propose a novel Mixture of Disentangled Prototypes (MoDP) method to im-prove the compositional generalization in CEL. The key idea is to introduce a finite prototype space that bridges the gap between unbounded agent-environment combinations and a single policy model. Specifically, we design a dual-headed autoencoder and a compositional reconstruction loss to disentangle agent and envi-ronment features from [...] 2 Related Works 2.1 Cross-Embodied Learning Cross-Embodiment Learning focuses on enabling robots to gen-eralize learned policies across different embodiments. Early work explored techniques such as conditioning on explicit representa-tions of the embodiment , domain randomization and adaptation [12, 34], and modular policies [19, 52], which were applied to sim-pler scenarios like single-task manipulations. Subsequent research has focused on alignment methods to develop universal strategies 7163 Improving Compositional Generalization in Cross-Embodiment Learning via Mixture of Disentangled Prototypes MM ’25, October 27–31, 2025, Dublin, Ireland that can transfer across tasks, such as manipulation and navigation. [...] 1 Introduction Cross-Embodied Learning [10, 50] (CEL) seeks to train a unified policy applicable to heterogeneous robot entities and capable of interacting with various environments, leading to better robustness and generalization. The resulting well-trained model, known as the generalist embodied agent , is expected to drive advancements in fields such as home service, autonomous driving, and multi-agent collaboration [11, 60]. As collecting embodied data from a single robot embodiment remains costly and inefficient, CEL has become an increasingly prominent research focus.

• [PDF] Cross-Embodiment Zero-Shot Policy Transfer with Cross-Painting

  We present Mirage (Fig. 1), a novel cross-embodiment policy transfer method that can zero-shot transfer policies trained on one source robot to an unseen target robot. Mirage decouples vision and control, allowing gaps between the source and target robot to be addressed separately. To address visual differences, Mirage employs “cross-painting” during execution, masking out the target robot and inpainting the source robot at the same pose, creating a “mirage” for the policy. Importantly, Mirage only requires knowledge of robot base coordinate frames for visual and state alignment. Mirage does not assume any demonstration data on the target robot or paired images or trajectories between the source and unseen target robots. Moreover, Mirage applies to both first-person and third-person [...] Robotics: Science and Systems 2024 Delft, Netherlands, July 15-July 19, 2024 Mirage: Cross-Embodiment Zero-Shot Policy Transfer with Cross-Painting Lawrence Yunliang Chen∗1, Kush Hari∗1, Karthik Dharmarajan∗1, Chenfeng Xu1, Quan Vuong2, Ken Goldberg1 1 UC Berkeley 2 Google DeepMind Fig. 1: Overview of Mirage. We study zero-shot policy transfer across embodiments. Assume there is a policy trained on a source robot (left). At test time, with an unseen target robot (middle), Mirage performs “cross-painting”—masking out the target robot in the image and inpainting the source robot at the same end effector pose—using robot URDFs and a renderer. By creating an illusion as if the source robot were performing the task (right), Mirage queries the source policy with the cross-painted image to