KITE: Decoupling Kinematics and Interaction for Zero-Shot Cross-Embodiment Manipulation

2026-06-20

AI Summary

This work addresses the challenge of zero-shot policy transfer across heterogeneous robot embodiments by proposing the KITE framework, which decouples manipulation tasks into embodiment-agnostic task reasoning and embodiment-specific motion control. The two components are bridged through a latent representation of interaction intent learned from contact patterns. Requiring only the target embodiment’s kinematic model and no additional task demonstration data, KITE enables efficient cross-embodiment transfer. Experimental results demonstrate that KITE significantly outperforms existing methods across three representative manipulation tasks involving parallel-jaw grippers, dexterous hands, and composite embodiments, achieving notable improvements in both transfer success rate and generalization capability.

Abstract

Generalizing manipulation policies across robot embodiments remains difficult because standard policies entangle task reasoning with embodiment-specific motor control. We study zero-shot cross-embodiment manipulation, where a policy trained on source embodiments must be deployed on a structurally different target embodiment without additional task demonstrations. We introduce Kinematic Interaction Transfer across Embodiments (KITE), which decouples manipulation into embodiment-agnostic task reasoning and embodiment-specific motor control, connected through a learned latent representation of interaction intent based on contact patterns. Task reasoning is performed by a shared policy that predicts latent intents from source demonstrations, while motor control is performed by an intent-conditioned action decoder learned from each embodiment's kinematic model. With KITE, adaptation to a new embodiment requires only training a new action decoder using its kinematic model, without recollecting demonstration data. We evaluate KITE on three manipulation tasks spanning transfer between parallel grippers, dexterous hands, and composite embodiments. KITE consistently achieves zero-shot transfer to structurally different target embodiments, outperforming state-of-the-art baselines in transfer success and task-embodiment scope.

Problem

• zero-shot
• cross-embodiment
• manipulation
• generalization
• robot embodiment

Innovation

• zero-shot transfer
• cross-embodiment manipulation
• kinematic decoupling
• latent interaction intent
• embodiment-agnostic policy