Frequently Asked Questions

Kyber Labs is using our novel artificial muscle technology to build a low cost robotic manipulation platform designed for ML controls. Our muscle tech was inspired by human muscle similar to how neural nets were inspired by the brain. Combining these two biomimetic approaches allows us to create a hardware system that's as versatile as the software and enables embodied AI to interact with the world fluidly and naturally. This capability will allow us to automate tasks that are currently intractable and accelerate human productivity. 

With the advent of ML models that can comprehend the world and act on it intelligently we see a huge gap in the capabilities of current hardware to make use of that intelligence. Conventional motors / gearboxes are reaching their optimization limit and were fundamentally not designed for ML controls, leading to only marginal improvements in cost and capability. Thus we believe a paradigm shift is necessary to support the rising diversity of use cases enabled by AI.

The product we are working towards is a general purpose robotic manipulation platform that is designed to mimic the capabilities of the upper arms and hands of a normal human but for a fraction of the cost. This can then be adapted to a variety of applications and use cases beyond what robots are used for today. It can be stationary and do assembly and fabrication tasks for the logistics or manufacturing sectors with minimal setup. It can be put on a mobile base to do work around flat facilities like warehouses or put on a legged base to build a full humanoid robot. The core theme in these products is robots that can manipulate objects usefully in the real world alongside humans.

This is a technically challenging product to build and we plan to take incremental steps to get there with earlier products that are aligned with this long term goal but are easier to build and deploy. We're currently in stealth mode as we continue to develop our core product. 

Our artificial muscle technology represents a significant paradigm shift in robotics design, offering advantages uniquely aligned with ML-based controls:

• Low cost: Muscle is made from inexpensive materials, assembled from raw materials in-house, and can be universally scaled to any robot joint. We only need to change how many are bundled together and their length which reduces unique parts and manufacturing complexity.
• Robustness: The fully backdriveable design interacts only electromagnetically, ensuring no damage even from high external impulse forces.
• High efficiency: The muscles use electromagnetic interactions to produce force and approach the high mechanical efficiency of standard electric motors.
• High force: Our muscle fibers are stronger than human muscle per volume and thus building a human strength robot arm would be possible with a similar form factor. 
• Stiffness Control: Enables simultaneous joint position and stiffness control via antagonistic actuation. This is how biological muscles work and is crucial for human-like interaction with the real world. It eliminates the need to explicitly calculate every joint angle, relying on the system's inherent compliance to handle uncertainty.
• Inherent sensors: Being entirely force driven, the fiber actuators can forward calculate force based solely on the current in instead of back calculating the force from sensors. The fibers also know their contraction position which can be easily read by any microcontroller. 
• Simple controls: No fluid pressure systems or high voltages necessary, simply low voltage electric signal similar to conventional brushed motors.
• Dexterity: Made of small, flexible fibers, these actuators readily enable high dexterity systems with many degrees of freedom, outperforming traditional methods for complex tasks.
• Common Training Data for ML: Directly mimicking human interaction unlocks a vast training set of human video data, facilitating ML algorithm training.

The field of artificial muscle research has many different  avenues that are being explored but after following the field for a long time and researching the current options we found all of them to be significantly lacking in at least one key area. So we set out to build our own that is a completely new design based on first principles that is more balanced and designed with commercial viability as the driving factor.