YC Startup 'Origami Robotics' Tackles Dexterous Hands
Y Combinator has launched Origami Robotics, a startup developing high-degree-of-freedom robotic hands. The company is using in-joint motors and a unique data-collection glove to capture real-world data, aiming to solve one of the biggest challenges in embodied AI: dexterous manipulation.
The company's founders are brothers Daniel and Ryan Xie. Daniel's PhD research in manipulation at Carnegie Mellon's Robotics Institute repeatedly hit a wall due to the lack of robotic hands designed for learning and the scarcity of high-quality manipulation data. This firsthand frustration with existing hardware limitations, particularly the modeling difficulties caused by large transmission systems in conventional robotic hands, directly led to the founding of Origami Robotics with his brother Ryan, a serial robotics entrepreneur. Origami Robotics' core technical approach is the co-design of its high-degree-of-freedom (DOF) hand and a corresponding data-collection glove. This strategy directly targets the "embodiment gap," the significant differences between a human hand and a robotic hand in kinematics and sensing, which has been a persistent obstacle in robotics research. By ensuring the glove and hand match exactly, the company aims to collect highly transferable, real-world data for training manipulation policies. The startup's hands utilize in-joint, direct-drive motors, a key design choice for enabling advanced learning and control. Unlike geared motors, direct-drive systems eliminate backlash (the "play" in gears), increase precision, and allow for more transparent force feedback, which is critical for tasks requiring fine-touch sensing. This design simplifies the hand's dynamics, making it easier to model and more suitable for transferring skills learned in simulation to the real world. This focus on data and hardware synergy places Origami Robotics at the intersection of two major trends: embodied AI and foundation models. Embodied AI posits that intelligence is developed through physical interaction with the world. By creating a system to scale the collection of real-world manipulation data, Origami is building the infrastructure needed to train large-scale, generalizable "manipulate anything" models for robotics. The Y Combinator-backed startup has already sold its hardware to major players like Amazon's Physical AI labs to push the frontiers of manipulation research. This early adoption by industry leaders highlights the demand for better hardware to unlock the potential of AI in robotics. The company's strategy is to deploy its devices in manufacturing and logistics centers to gather massive, Tesla-style datasets to train and refine its AI models. The competitive landscape for dexterous hands includes companies like Mimic, Tesollo, and China-based LinkerBot. However, many competitors focus on the hardware alone. Origami's differentiation lies in its integrated approach, tightly coupling the hardware design with a data-collection strategy specifically aimed at solving the learning problem for general-purpose manipulation.
