LEARNINGCURRENT2026-06-15

Unified Motion-Action Modeling for Heterogeneous Robot Learning

Yunhao Cao, Shitong Liu, Chao Feng, Meryl Zhang, Xuanchen Lu, Andrew Owens, Kuan Fang

This work unifies visuomotor Control & PlanningControlThe method used to make the robot move the way you want. and Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. modeling by using 3D object motion as a shared representation, letting a single pretrained model handle motion-conditioned Control & PlanningControlThe method used to make the robot move the way you want., forward Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. prediction, and Modern Robot LearningFew-shotLearning a new task from only a small number of examples. Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. learning without task-specific retraining. The key insight: treating Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions and object trajectories as co-evolving variables with masked generative modeling eliminates the need for manual Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. annotations when combining Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. demos, human videos, and Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. data.

THE PROBLEM

This paper focuses on learning. UMA uses 3D object motion trajectories as a unified interface between visuomotor Control & PlanningControlThe method used to make the robot move the way you want. and forward Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia.. A masked generative model with contrastive disentanglement enables multi-task Modern Robot LearningPretrainingTraining a model on a broad dataset before adapting it to a specific task. across heterogeneous data (Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions., human, sim) without manual labels. The same pretrained model supports three modes at Simulation & Sim-to-RealDeploymentPutting the trained system on a real robot.: motion-conditioned Control & PlanningControlThe method used to make the robot move the way you want., Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. prediction, and Modern Robot LearningFew-shotLearning a new task from only a small number of examples. adaptation. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.

HOW IT WORKS

1

Task framing

The paper frames the work as learning. Start here because it defines what success means and which assumptions the rest of the method inherits.

2

Core method

This work unifies visuomotor Control & PlanningControlThe method used to make the robot move the way you want. and Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. modeling by using 3D object motion as a shared representation, letting a single pretrained model handle motion-conditioned Control & PlanningControlThe method used to make the robot move the way you want., forward Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. prediction, and Modern Robot LearningFew-shotLearning a new task from only a small number of examples. Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. learning without task-specific retraining. The key insight: treating Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions and object trajectories as co-evolving variables with masked generative modeling eliminates the need for manual Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. annotations when combining Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. demos, human videos, and Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. data. When reading the method section, identify the inputs, the learned or engineered representation, and the Core ConceptsActionA command the robot sends to its motors, controller, or low-level system. or prediction produced by the system.

3

Data and supervision

For robotics work, the data story is part of the method: check whether the system depends on Imitation & Reinforcement LearningTeleoperation (teleop)A human remotely controlling the robot, often to collect demonstrations., Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested., internet video, human labels, or Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. rollouts.

4

Evaluation evidence

The paper should be judged through its Simulation & Sim-to-RealEvaluationMeasuring how well a robot system performs. protocol: what data is used, what Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or simulator is tested, and which Evaluation & ResearchBaselineA reference method used for comparison. comparisons support the claim. Look for the gap between the headline result and the Simulation & Sim-to-RealDeploymentPutting the trained system on a real robot. setting you would actually care about.

KEY RESULTS

Main contributionConceptual contribution

This work unifies visuomotor Control & PlanningControlThe method used to make the robot move the way you want. and Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. modeling by using 3D object motion as a shared representation, letting a single pretrained model handle motion-conditioned Control & PlanningControlThe method used to make the robot move the way you want., forward Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. prediction, and Modern Robot LearningFew-shotLearning a new task from only a small number of examples. Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. learning without task-specific retraining. The key insight: treating Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions and object trajectories as co-evolving variables with masked generative modeling eliminates the need for manual Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. annotations when combining Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. demos, human videos, and Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. data.

WHY DEVELOPERS SHOULD CARE

This work unifies visuomotor Control & PlanningControlThe method used to make the robot move the way you want. and Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. modeling by using 3D object motion as a shared representation, letting a single pretrained model handle motion-conditioned Control & PlanningControlThe method used to make the robot move the way you want., forward Movement, Mechanics & Robot BodyDynamicsThe study of motion including forces, torques, mass, and inertia. prediction, and Modern Robot LearningFew-shotLearning a new task from only a small number of examples. Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. learning without task-specific retraining. The key insight: treating Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions and object trajectories as co-evolving variables with masked generative modeling eliminates the need for manual Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. annotations when combining Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. demos, human videos, and Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. data.

LIMITATIONS

The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.

WHAT COMES NEXT

The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's learning assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.

RELATED PAPERS