Ask Onix
Human hands: A marvel of biological engineering
The human hand stands as one of the body's most complex structures, boasting over 30 muscles, 27 joints, and 27 degrees of freedom. Its network of ligaments, tendons, and 17,000 touch receptors in the palm alone enables an extraordinary range of movements-from delicate tasks like writing to powerful grips.
For Sarah de Lagarde, this complexity became painfully evident after a 2022 accident at London's High Barnet station left her without her right arm below the shoulder. A standard NHS prosthetic offered little functionality, leaving her struggling with daily tasks. "It doesn't really look like a real arm," she recalls. "It was deemed creepy by my children."
A new era of bionic limbs
De Lagarde's fortunes changed when she received an AI-driven bionic arm. Unlike static prosthetics, this device uses machine learning to interpret electrical signals from her muscles, predicting movements in real time. "Every time I make a movement, it learns," she explains. "Eventually, it starts predicting what my next move is."
The arm's generative AI adapts to her intentions, allowing her to adjust grip strength-from holding an egg without crushing it to compressing a soda can. An accompanying app even suggests optimizations based on her usage patterns. Yet challenges remain: the prosthesis lacks advanced haptic feedback, forcing De Lagarde to rely on sight to avoid dropping objects.
Robots learn through trial and error
Replicating human dexterity has long eluded engineers. Early attempts, like 16th-century iron hands, pale in comparison to modern AI-powered robots. Today, systems like Google DeepMind's DEX-EE-a three-fingered robotic hand with 12 degrees of freedom-can manipulate fragile objects, even learning to shake hands through reinforcement learning.
"Traditional AI processes information, but embodied AI perceives and reacts to the physical world," says Eric Jing Du, a civil engineering professor at the University of Florida. His team develops robots that refine movements through real-world interactions, much like a baby learning to grasp objects.
From farms to nuclear sites: Robots in action
While research tools like DEX-EE dominate labs, practical applications are emerging. Dogtooth Technologies' fruit-picking robots use machine vision and seven-degree-of-freedom arms to harvest strawberries without damage. Meanwhile, Rustam Stolkin's team at the University of Birmingham designs robots to handle hazardous nuclear waste, where human presence is dangerous.
Boston Dynamics' Atlas robot exemplifies progress, combining computer vision and reinforcement learning to perform tasks like shelf organization. Yet, as Du notes, "Most robots are trained for specific tasks and struggle with unpredictability."
Barriers to human-like dexterity
Despite advances, robots lag behind human hands in adaptability. A recent test of a $5,000 robotic hand revealed it dropped a rubber duck 56% of the time. Tesla's Optimus robot, though boasting 25 degrees of freedom, still relies on remote control for tasks like catching tennis balls.
"Human sensory systems detect minute changes and integrate multiple inputs," Du explains. "Robots lack this level of perception." Safety and ethics also pose hurdles, from job displacement to ensuring robots don't harm human coworkers.
A future of collaboration
For De Lagarde, bionic advancements have restored lost abilities-pouring water, hugging her children. She envisions a future where robotic augmentation aids the elderly and disabled. "I can instruct it to have a very light touch," she says, highlighting the arm's precision.
Pulkit Agrawal of MIT predicts transformative impacts across industries, from manufacturing to elder care. Yet, he cautions, "Human-like dexterity is at least five years away."
"Whether in prosthetics or robotics, AI is unlocking capabilities we once thought impossible," says Agrawal. "But we're still chasing the extraordinary design of the human hand."
