GRASPINGCURRENT2026-04-14

Learning Versatile Humanoid Manipulation with Touch Dreaming

Yaru Niu, Zhenlong Fang, Binghong Chen, Shuai Zhou, Revanth Senthilkumaran, Hao Zhang, Bingqing Chen, Chen Qiu, H. Eric Tseng, Jonathan Francis, Ding Zhao

ARCHITECTURE
multimodal encoder-decoder Transformer with touch dreaming
ROBOT
humanoid
KEY METRIC
90.9%
TASK
manipulation, loco-manipulation

This paper shows how to make humanoid robots do complex Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks (folding towels, organizing books) by treating touch/force Control & PlanningFeedbackInformation returned from sensors during action to help correct behavior. as a first-class prediction target alongside vision. Instead of just predicting next actions, the Core ConceptsPolicyThe rule or model that maps observations or states to actions. also predicts future tactile sensations, which forces the model to learn contact-aware representations—achieving 90% better success rates than vision-only baselines on contact-rich tasks.

THE PROBLEM

This paper focuses on Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects., Manipulation & TasksLoco-manipulationLocomotion and manipulation happening together, often in humanoids.. This paper shows how to make humanoid robots do complex Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks (folding towels, organizing books) by treating touch/force Control & PlanningFeedbackInformation returned from sensors during action to help correct behavior. as a first-class prediction target alongside vision. Instead of just predicting next actions, the Core ConceptsPolicyThe rule or model that maps observations or states to actions. also predicts future tactile sensations, which forces the model to learn contact-aware representations—achieving 90% better success rates than vision-only baselines on contact-rich tasks. 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 Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects., Manipulation & TasksLoco-manipulationLocomotion and manipulation happening together, often in humanoids.. The reported platform or hardware context is humanoid. The Simulation & Sim-to-RealEvaluationMeasuring how well a robot system performs. setting is real-world testing. Start here because it defines what success means and which assumptions the rest of the method inherits.

2

Core method

The method is organized around Modern Robot LearningMultimodalUsing more than one type of input, like vision, language, touch, or proprioception. encoder-decoder Transformer with touch dreaming. This paper shows how to make humanoid robots do complex Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks (folding towels, organizing books) by treating touch/force Control & PlanningFeedbackInformation returned from sensors during action to help correct behavior. as a first-class prediction target alongside vision. Instead of just predicting next actions, the Core ConceptsPolicyThe rule or model that maps observations or states to actions. also predicts future tactile sensations, which forces the model to learn contact-aware representations—achieving 90% better success rates than vision-only baselines on contact-rich tasks. 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 key reported result is HTD achieves 90.9% relative improvement in average Simulation & Sim-to-RealSuccess rateHow often the robot completes a task correctly. over Evaluation & ResearchBaselineA reference method used for comparison. across five contact-rich Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks 90.9%. 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

Primary metric90.9%

HTD achieves 90.9% relative improvement in average Simulation & Sim-to-RealSuccess rateHow often the robot completes a task correctly. over Evaluation & ResearchBaselineA reference method used for comparison. across five contact-rich Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks

WHY DEVELOPERS SHOULD CARE

This paper shows how to make humanoid robots do complex Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects. tasks (folding towels, organizing books) by treating touch/force Control & PlanningFeedbackInformation returned from sensors during action to help correct behavior. as a first-class prediction target alongside vision. Instead of just predicting next actions, the Core ConceptsPolicyThe rule or model that maps observations or states to actions. also predicts future tactile sensations, which forces the model to learn contact-aware representations—achieving 90% better success rates than vision-only baselines on contact-rich tasks.

LIMITATIONS

The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing beyond humanoid.

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 Manipulation & TasksManipulationUsing a robot arm or hand to move or interact with objects., Manipulation & TasksLoco-manipulationLocomotion and manipulation happening together, often in humanoids. 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