US-based Foundation Robotics has unveiled a new video showing its latest robotic hand cleanly catching a baseball in mid-flight, highlighting major advances in robotic dexterity and control.

The tendon-driven hand uses motors housed in the forearm instead of the fingers, creating a slimmer, lighter design while maintaining precise movements.

Its fingers automatically adjust their shape to securely grip round objects like a baseball, improving both speed and reliability during catches.

According to the San-Francisco-based firm, the demonstration highlights how improved mechanical design and motion planning are bringing robotic hands closer to performing fast, complex tasks with human-like precision.

Robotic hand evolution

Foundation has demonstrated the capabilities of the new generation of its robotic hands by making it catch a baseball in mid-flight—a task that demands fast motion, precise timing, and reliable gripping.

While the throw is carefully planned in advance, the demonstration showcases the hand’s advanced mechanical design and control system, highlighting how modern robotics is moving closer to human-like dexterity for industrial applications.

At the heart of the new design is a tendon-driven architecture that relocates the motors from the fingers into the robot’s forearm. This approach significantly reduces the weight and bulk of the fingers, allowing them to remain slim while still generating the force needed for rapid, precise movements. Flexible tendons route from the forearm motors through carefully engineered pathways to each finger joint. Separate flexion tendons close the fingers, while extension tendons reopen them, enabling smooth, coordinated motion similar to the tendons in a human hand.

The team also developed a mechanism that allows subtle side-to-side finger movements, enabling the hand to change shape depending on the object it is handling. Instead of simply opening and closing, the fingers can form a cupped shape to securely cradle spherical objects such as a baseball or adjust into more precise pinch grips for smaller items. This adaptability expands the range of objects the hand can manipulate without requiring specialized grippers.

Adaptive robotic grasping

A major technical breakthrough lies in the hand’s ability to estimate the position of every finger without depending entirely on physical joint sensors. Foundation’s control software continuously calculates finger positions by combining motor rotation data with a detailed model of the tendons’ geometry. This software-based estimation provides real-time awareness of the hand’s configuration, allowing it to continue operating even if individual sensors become unavailable.

For added accuracy, each joint is also equipped with tunnel magnetoresistance (TMR) sensors that measure joint angles with sub-degree precision. Rather than serving as the primary source of position data, these sensors refine the software’s estimates and provide redundancy, improving reliability in demanding industrial environments where sensors may occasionally fail or become damaged.

The baseball-catching demonstration also highlights the importance of minimizing friction within the tendon system. Low-friction routing ensures motor movements translate accurately into finger motion, maintaining synchronization between commanded and actual positions. That precision allows the fingers to close quickly while absorbing the ball’s impact without bouncing it away, reports Techeblog.

The robotic hand is intended for Foundation’s Phantom humanoid robots, which are designed for industrial workplaces. Unlike the company’s earlier gripper-style hands, the new prototype features independently actuated fingers, anatomically inspired joints, advanced state estimation, and adaptive grasping. Together, these technologies enable the robot to manipulate a much wider variety of tools, components, and irregular objects, bringing humanoid robots closer to performing complex real-world tasks with greater speed, precision, and reliability.