Party OS

An open-source R&D foundation for humanoid robotics,built for developers worldwide

GitHub

Explore the documents

We empower every mind with talent, craft, and vision to be seen by the world. Here, people driven by pure passion come together to build things that are boldly frontier, deeply real, and truly meaningful.

Party OS

Locomotion

Build locomotion infrastructure spanning data generation, motion retargeting, training frameworks, Sim2Real / Real2Sim deployment, and Behavior Foundation Models (BFM) to advance robot motion capabilities and development efficiency.

Slide 1 of 6

UFO
An open-source unsupervised reinforcement learning framework for robot control. Beyond FB, UFO connects training with deployment on real robots.
MimicLite
A fully open-source supervised learning framework for dynamic motion tracking and low-latency teleoperation. Trained on 8 GPUs in 3 hours.
Human-Humanoid Tools
Retarget complex whole-body motions in about 30 seconds. Supports different humanoid structures, batch processing, motion analysis, and robot-to-robot motion transfer.
I-BFM
An open-source behavior foundation model for humanoid–object interaction. One policy uses unsupervised learning to carry, push, and kick objects without task-specific retraining.
P-BFM
A perceptive behavior foundation model for whole-body terrain control. One policy combines depth perception, motion references, and commands for zero-shot real-world deployment.
PredActor
An open-source humanoid diffusion policy that jointly predicts whole-body states and joint actions, running at 50 Hz on Jetson Orin NX without a separate tracker.
UFO
An open-source unsupervised reinforcement learning framework for robot control. Beyond FB, UFO connects training with deployment on real robots.
MimicLite
A fully open-source supervised learning framework for dynamic motion tracking and low-latency teleoperation. Trained on 8 GPUs in 3 hours.
Human-Humanoid Tools
Retarget complex whole-body motions in about 30 seconds. Supports different humanoid structures, batch processing, motion analysis, and robot-to-robot motion transfer.
I-BFM
An open-source behavior foundation model for humanoid–object interaction. One policy uses unsupervised learning to carry, push, and kick objects without task-specific retraining.
P-BFM
A perceptive behavior foundation model for whole-body terrain control. One policy combines depth perception, motion references, and commands for zero-shot real-world deployment.
PredActor
An open-source humanoid diffusion policy that jointly predicts whole-body states and joint actions, running at 50 Hz on Jetson Orin NX without a separate tracker.
UFO
An open-source unsupervised reinforcement learning framework for robot control. Beyond FB, UFO connects training with deployment on real robots.
MimicLite
A fully open-source supervised learning framework for dynamic motion tracking and low-latency teleoperation. Trained on 8 GPUs in 3 hours.
Human-Humanoid Tools
Retarget complex whole-body motions in about 30 seconds. Supports different humanoid structures, batch processing, motion analysis, and robot-to-robot motion transfer.
I-BFM
An open-source behavior foundation model for humanoid–object interaction. One policy uses unsupervised learning to carry, push, and kick objects without task-specific retraining.
P-BFM
A perceptive behavior foundation model for whole-body terrain control. One policy combines depth perception, motion references, and commands for zero-shot real-world deployment.
PredActor
An open-source humanoid diffusion policy that jointly predicts whole-body states and joint actions, running at 50 Hz on Jetson Orin NX without a separate tracker.

Perception

Develop environmental perception, human–scene interaction (HSI), and human–object interaction (HOI) to strengthen perception, decision-making, and continuous interaction in complex real-world environments, laying a stable foundation for manipulation intelligence.

Slide 1 of 3

Environmental Perception
Focus on perceiving complex real-world environments to provide a stable foundation for robot decision-making and continuous interaction.
Human–Scene Interaction
Explore human–scene interaction (HSI) to improve how robots perceive and interact with their surroundings in complex real-world environments.
Human–Object Interaction
Explore human–object interaction (HOI) as part of the perception and interaction capabilities that support future manipulation intelligence.
Environmental Perception
Focus on perceiving complex real-world environments to provide a stable foundation for robot decision-making and continuous interaction.
Human–Scene Interaction
Explore human–scene interaction (HSI) to improve how robots perceive and interact with their surroundings in complex real-world environments.
Human–Object Interaction
Explore human–object interaction (HOI) as part of the perception and interaction capabilities that support future manipulation intelligence.
Environmental Perception
Focus on perceiving complex real-world environments to provide a stable foundation for robot decision-making and continuous interaction.
Human–Scene Interaction
Explore human–scene interaction (HSI) to improve how robots perceive and interact with their surroundings in complex real-world environments.
Human–Object Interaction
Explore human–object interaction (HOI) as part of the perception and interaction capabilities that support future manipulation intelligence.

Manipulation

Bring together Behavior Foundation Models (BFM), vision-language-action models (VLA), and World Models to advance whole-body coordination and complex manipulation, building an infrastructure for manipulation intelligence that can generalize at scale.

Slide 1 of 3

Behavior Foundation Models
Integrate BFM capabilities into the development of whole-body coordination and complex manipulation for humanoid robots.
Vision-Language-Action Models
Combine VLA capabilities with motion foundations to advance coordinated whole-body control and complex manipulation tasks.
INTACT
Explore World Models alongside BFM and VLA as part of an infrastructure for manipulation intelligence with scalable generalization.
Behavior Foundation Models
Integrate BFM capabilities into the development of whole-body coordination and complex manipulation for humanoid robots.
Vision-Language-Action Models
Combine VLA capabilities with motion foundations to advance coordinated whole-body control and complex manipulation tasks.
INTACT
Explore World Models alongside BFM and VLA as part of an infrastructure for manipulation intelligence with scalable generalization.
Behavior Foundation Models
Integrate BFM capabilities into the development of whole-body coordination and complex manipulation for humanoid robots.
Vision-Language-Action Models
Combine VLA capabilities with motion foundations to advance coordinated whole-body control and complex manipulation tasks.
INTACT
Explore World Models alongside BFM and VLA as part of an infrastructure for manipulation intelligence with scalable generalization.

Agentic Humanoid

Build on an Agent + Skills architecture to explore robot task planning, autonomous execution, and skill composition, advancing humanoid robots from single-task execution toward autonomous agents.

Slide 1 of 3

Task Planning
Explore robot task planning through an Agent + Skills architecture, supporting the transition from isolated tasks toward autonomous agents.
Autonomous Execution
Investigate autonomous execution as a core capability of humanoid agent systems built on Agent + Skills.
Skill Composition
Explore how skills can be composed within an agent system to move humanoid robots beyond single-task execution.
Task Planning
Explore robot task planning through an Agent + Skills architecture, supporting the transition from isolated tasks toward autonomous agents.
Autonomous Execution
Investigate autonomous execution as a core capability of humanoid agent systems built on Agent + Skills.
Skill Composition
Explore how skills can be composed within an agent system to move humanoid robots beyond single-task execution.
Task Planning
Explore robot task planning through an Agent + Skills architecture, supporting the transition from isolated tasks toward autonomous agents.
Autonomous Execution
Investigate autonomous execution as a core capability of humanoid agent systems built on Agent + Skills.
Skill Composition
Explore how skills can be composed within an agent system to move humanoid robots beyond single-task execution.

Agentic RSI System

Supporting the new generation of physical AI

Our mission is to empower high potential, early stage startups through Google's AI and Gemini Robotics models. We're helping transform their discoveries into market-leading ventures – and to power an era of intelligent physical AI.

Learn more

Join RoboParty Lab

We welcome researchers, engineers, and open-source contributors who care about humanoid robotics. Explore locomotion, perception, and manipulation with us, turn ideas into capabilities on real robots, and share what we build with developers everywhere. Pursue the frontier. Stay true to what drives us.

Learn more