Origami Robotics

Roadmap & Position in Robotics Manipulation

Builds robotic hands with matched data-collection gloves for smooth human-to-robot transfer.

Company Overview

Builds high-degree-of-freedom robotic hands with co-designed data-collection gloves that enable general-purpose dexterous manipulation through scalable sim-to-real transfer and learning-based policies.

What They're Building

The company's public product roadmap & what they're committed to building.

Origami Robotics has publicly showcased their high-DOF robotic hand with in-joint motors and a kinematically matched data-collection glove designed for smooth sim-to-real policy transfer. They have announced early commercial sales to Amazon's Physical AI Labs and are part of Y Combinator's Winter 2026 batch, signaling a focus on scaling hardware production and expanding their manipulation data infrastructure.

Latest Intelligence

Zeitgeist tracks private signals to determine where the company is heading strategically.

Hands Become The Manipulation Data Layer

May 13, 2026

Confidence:

High

New Intel: Origami Robotics is pairing robotic hands with matched data gloves for sim-to-real manipulation. Amazon Physical AI Labs buying early points to a path from hardware into manipulation data infrastructure.

Founder and Key Execs

Quanting (Daniel) Xie, Co-Founder

Co-Founder

Ryan Xie, Co-Founder

Co-Founder

Founder Force Multiplier

Origami Robotics uniquely co-designs their robotic hand and data-collection glove with identical kinematics and sensing, eliminating the embodiment gap that plagues competitors and enabling direct, high-fidelity policy transfer from human demonstration to robot, a hardware moat.

Funding History

2025 | Quanting Xie researches dexterous manipulation at CMU Robotics Institute.
2026 | Origami Robotics founded.
2026 | Accepted into Y Combinator W26 batch.
2026 | Early commercial sales to Amazon's Physical AI Labs.

Competitors

Dexterous Hand Hardware

Shadow Robot (Dexterous Hand), Wonik Robotics (Allegro Hand), Leap Hand (CMU open-source).

Learning-Based Manipulation

Tesla Optimus (in-house hands), Figure AI (Figure 02), Sanctuary AI (Phoenix).

Data & Teleoperation

DexCap (Stanford), AnyTeleop, Haptx (haptic gloves).

Foundation Model Robotics

Physical Intelligence (Pi), Covariant (now part of Amazon), Skild AI.

Origami Robotics

's Moat:

Co-designed robotic hand and data-collection glove with identical kinematics and sensing. This eliminates the embodiment gap that makes other systems' policy transfer lossy. Amazon Physical AI Labs as an early customer validates the hardware. The co-design approach is a mechanical engineering moat that software-only competitors cannot replicate.

How They're Leveraging AI

Sim-to-Real Imitation Learning

Sim-to-Real Dexterous Manipulation Policy Transfer: Uses imitation learning from human demonstrations captured via a kinematically matched data-collection glove to train and deploy dexterous manipulation policies directly on the robotic hand with minimal domain adaptation.

Large-Scale Manipulation Data Pipeline

Scalable Manipulation Data Engine: Deploys fleets of data-collection gloves to crowdsource thousands of hours of diverse, real-world manipulation demonstrations, building a proprietary dataset that trains increasingly generalizable foundation models for robotic dexterity.

Dynamic Dexterity via RL

Reinforcement Learning for Dynamic In-Hand Manipulation: Trains RL agents in high-fidelity simulation to perform complex dynamic manipulation tasks (e.g., pen spinning, object reorientation, tool pivoting) and deploys them on the physical hand, using the accurate joint-level actuation model enabled by in-joint motors.

AI Use Overview:

Using sim-to-real imitation learning from kinematically matched demonstrations, large-scale manipulation data pipelines, and dynamic dexterity via RL.

More Similar Companies

Bedrock Robotics

Retrofit autonomy kits that convert excavators and heavy equipment into operator-less machines.

Ex-Waymo trucking leadership attacking a labor-starved $13T construction market with reversible retrofits instead of new OEM machines, a faster path to revenue than highway autonomy.

Waabi

AI-first autonomy stack for driverless trucks and robotaxis, validated in generative simulation.

Simulation-first development with generative world models lets Waabi validate safety without the fleet burn that bankrupted earlier AV companies, arriving at driverless launch with less capital consumed.

World Labs

Generates navigable 3D worlds from text and images using large world models.

The founders invented ImageNet and NeRF, giving them training data and architecture intuition in 3D generation that well-funded competitors have to rediscover from scratch.

Wayve

End-to-end deep learning software for autonomous driving, licensed to OEMs and fleets

OEM-licensable embodied AI with real production wins positions Wayve as the neutral AV software layer while Tesla and Waymo stay vertically integrated.

Back To All Companies >