Kyber Labs Emerges with a High-Speed, "Backdrivable" Robotic Hand

A new startup has quietly entered the increasingly crowded field of robotic manipulation, pitching a hardware-first approach to the industry's biggest bottleneck: dexterity.

Kyber Labs, a Brooklyn-based company founded by alumni from SpaceX and electric vehicle maker Tarform, broke cover on social media this month. While their public presence is currently sparse—consisting of a minimal website and a handful of posts on X (formerly Twitter)—the demonstrations they have shared are already turning heads among roboticists.

In a video posted recently, the company showed a blue, skeletal robotic hand spinning a nut onto a bolt at a speed that rivals human reflex. According to the company, the feat was performed "fully in real time" and demonstrates the hardware's core philosophy: backdrivability and torque transparency.

The "Soft" Hardware Approach

The current landscape of robotic hands is generally split between two extremes: rigid, high-precision industrial grippers that lack versatility, and complex, sensor-laden anthropomorphic hands that are often fragile and prohibitively expensive.

Kyber Labs appears to be targeting a middle ground, focusing on "compliance"—the ability of the robot to yield mechanically to its environment.

"Our hand rotates a nut on a bolt at super high speed... This is possible because it’s fully backdrivable and torque transparent so it adapts naturally to the nut," the company stated in a post.

In robotics, a "backdrivable" actuator allows an external force to move the motor. If you push on the robot's finger, it moves, and the computer knows it moved. This is in contrast to stiff, high-gear-ratio motors that lock in place. By monitoring the electrical current required to hold the finger in place, the robot can "feel" resistance without needing delicate tactile sensors on the fingertips.

To illustrate this, Kyber Labs shared a video and description of their hand interacting with a feather. The company claims the finger moves until it meets "tiny resistance" and stops instantly. Conversely, they claim the same mechanism allows the hand to "take a hammer hit" without breaking, a common failure point for rigid gearboxes.

A Different Path to Dexterity

This design philosophy places Kyber Labs in an interesting position relative to its competitors.

• Vs. Direct Drive: The emphasis on backdrivability echoes the "direct drive" approach seen in the Wuji Tech hand, which uses motors inside the fingers to minimize friction.
• Vs. Tactile Sensing: Unlike the SharpaWave hand, which relies on high-density pressure arrays to feel the world, Kyber is currently relying on "proprioception" (the robot's internal sense of position and effort). However, their FAQ notes they are developing "low cost, high density fingertip sensors" for the future.
• Vs. Cost: perhaps most significantly, Kyber Labs is aggressively targeting price. The company states their hardware is designed to cost "in the hundreds of dollars, not thousands." This puts them in the same disruptor category as, for example, the open-source RUKA hand, aiming to democratize access to high-end manipulation.

The Team and "Embodied AI"

While the hardware is the current focus, Kyber Labs positions itself as an "Embodied AI" company. Their thesis is that current hardware is too fragile and stiff to effectively train AI models using Reinforcement Learning (RL), which often requires robots to fail thousands of times in simulation or the real world.

The leadership team suggests a blend of aerospace engineering and academic robotics research:

• Tyler Habowski (Co-founder): A SpaceX veteran who worked on flight reusability systems for Starship and Falcon 9, and later as a Machine Learning Engineer at Machina Labs.
• Yonatan Robbins (Co-founder): An industrial designer formerly with Tarform, an electric motorcycle startup known for novel manufacturing.
• Julian Viereck (Robotics Research Scientist): A PhD from NYU who specialized in control software for locomotion and dexterity—a background that likely informs the "software-defined" nature of their hardware control.
• Lizzie Ayton (Mechanical Engineer): A Columbia trained engineer with experience in biotech and robotics.

Not Quite a Humanoid

Despite the anthropomorphic hand, Kyber Labs explicitly states in their FAQ that they are "not quite" making humanoid robots.

"We believe the true value of humanoid robots lies in their hands and what they can do with them in the world," the company writes. Their immediate goal is a "bimanual robotics platform" (two arms and hands) that can be mounted on various bases, targeting high-mix, low-volume industrial tasks that traditional automation misses.

This pragmatic approach mirrors a growing sentiment in the industry: while walking robots are impressive, the immediate economic value lies in manipulation. Companies like Boston Dynamics have similarly pivoted to emphasize the utility of their hands in their new electric Atlas platform.

Kyber Labs is currently backed by investors including Cortical Ventures, Starburst Ventures, and Trevor Blackwell. While their hardware appears to still be in the prototyping phase—evident from the 3D-printed components in their media—their focus on compliant, AI-ready, and low-cost hardware marks them as a startup to watch in 2026.