Experts Weigh In on Wuji Tech’s Robotic Hand: A “Remarkably Robust” Direct-Drive Contender

A new high-dexterity robotic hand from Shenzhen-based Wuji Tech, which we first covered last week, has captured the attention of the robotics community, not just for its performance specifications, but for its core design philosophy. In a recent technical breakdown on the "Over The Horizon" podcast, hosted by journalist Royden D'Souza, robotics expert Scott Walter and hand surgeon Gustav Andersson provided a detailed analysis, calling the hand "extremely robust" and a noteworthy challenger to the tendon-driven designs favored by competitors like Tesla.

The key innovation, as Walter and Andersson detailed, is Wuji's use of a direct-drive system, embedding micro-actuators directly within each finger phalanx. This stands in stark contrast to the more common approach of housing motors in the forearm and using a complex system of tendons to actuate the fingers.

A Different Approach to Actuation

"They're not using tendons," Walter stated plainly. Gustav Andersson elaborated, saying, "They're doing... the anti-version of Elon's approach of moving it all to the forearm... they'll keep it in the hand so there is no compromise."

This in-finger actuation provides direct, independent control over each of the hand's 20 joints. According to the expert panel, this design choice has significant advantages, particularly in bridging the gap between simulation and real-world performance. Tendon-based systems can be difficult to model accurately due to factors like tendon tension and stretching, leading to a "sim-to-real" gap. Wuji’s design, with its simple serial kinematics, is far more predictable.

"I absolutely agree with that," Walter said, referencing Wuji's claim of a minimal sim-to-real gap. "You don't have to worry about what's going on with tendons... They're basically like tiny little robots is what they've turned the fingers into."

Performance and Human-Like Form

The discussion highlighted the hand's impressive strength and dexterity. Promotional videos show the hand performing a 5kg pinch lift with just two fingers and a 20kg power grip, figures the experts found notable for a direct-drive system.

From an anatomical perspective, Gustav Andersson, a hand surgeon, noted that the Wuji Hand achieves a more natural palm shape than many of its counterparts. When clenched, the hand forms a realistic arch. However, he also pointed out areas where the design deviates from human anatomy.

"They seem to be going for a one-finger-length design," Andersson observed, which Walter agreed was likely for manufacturing simplicity. A human hand has fingers of varying lengths, which is critical for a perfectly closed grip.

The experts also analyzed the thumb's range of motion, a critical factor for dexterity. While the hand can touch the tip of the thumb to the little finger—a key metric in the Kapandji score of hand function—it does so with the side of the thumb's tip, not pulp-to-pulp. This, Andersson explained, is due to a lack of the subtle rotation a human thumb performs, limiting its ability to execute truly fine-motor tasks like tying a shoelace.

Open Questions and Future Potential

The panel noted the current absence of advanced tactile sensors in the hand's base model. While encoders provide positional feedback for the motors, there are no apparent sensors for pressure or texture. "For the first version of a hand, it's not too bad," Walter commented, speculating that sensors could be integrated into a future version of the glove-like skin.

Despite some limitations, the overall assessment was highly positive. The experts concluded that Wuji Tech's focus on a robust, direct-drive mechanical platform provides a powerful foundation. By solving the core actuation problem first, Wuji has created a hand that excels in predictability and control, even if it has yet to incorporate the full suite of sensory feedback seen in other prototypes.

"This is a really good first attempt," said Walter. "There's some room for improvement, but it's good... I still think that this would be a better approach to really being able to do sim-to-real and to emulate the small, fine motions."

Video