The robotics industry is shifting focus from how many fingers a robot hand has to whether it can actually complete complex tasks in the real world. On August 7, CASBOT Hands, a Beijing-based embodied intelligence company, introduced three new dexterous hand products and made a bold argument: counting degrees of freedom tells you almost nothing about whether a robot can reliably work in manufacturing, research, or service environments.

Why Robot Hands Matter More Than Ever?

Dexterous hands have become the critical bottleneck in humanoid robotics. These aren't just mechanical appendages; they're the primary interface between a robot and its physical environment. The market is growing explosively. In China alone, approximately 19,200 dexterous robotic hands were sold in 2025, a 236.84% increase year over year. Industry analysts project sales will reach approximately 70,200 units in 2026, representing 265.63% growth. By 2030, as humanoid robots become more common and costs decline, annual sales in China could exceed 430,000 units, with a compound annual growth rate of approximately 57.79% between 2026 and 2030.

This explosive growth reflects a fundamental shift in robotics. The industry is moving beyond proof-of-concept demonstrations toward real commercial deployment. Between 2023 and May 2026, approximately 57 financing transactions took place in China's dexterous hand sector, involving roughly 12.8 billion Chinese yuan in total investment.

What's Wrong With Comparing Robot Hands by Specifications Alone?

Most of the industry still evaluates dexterous hands using traditional metrics: degrees of freedom, weight, payload capacity, speed, precision, and sensor configurations. These numbers matter, but they miss something crucial. A robot with 20 degrees of freedom isn't automatically better than one with 15 if it can't coordinate those joints effectively. The real measure of dexterity isn't mechanical complexity; it's whether a robot can progress from simple grasping to actual manipulation and task completion.

"A robot that can reliably pick up a screwdriver has demonstrated grasping capability. Actually turning a screw requires control of the tool's orientation, position, trajectory, and applied force," explained CASBOT Hands in describing the gap between hardware specifications and real-world capability.

CASBOT Hands, Embodied Intelligence Company

This distinction matters because it explains why impressive lab demonstrations often fail in real commercial environments. Fine manipulation requires the coordinated interaction of mechanical structure, perception, force control, and accumulated task experience. A high degree of freedom, by itself, doesn't translate into useful manipulation skills.

How CASBOT Hands Is Approaching the Problem Differently

Rather than developing a single dexterous hand for all use cases, CASBOT Hands has adopted multiple technical approaches designed around different task requirements:

• The L1 Hand: A lightweight design with streamlined degrees of freedom, optimized for cost efficiency, stability, and high-frequency repetitive work. It was previously demonstrated at the World Artificial Intelligence Conference, where four robots equipped with the hand formed the CASBOT BAND and performed live music using real instruments, including guitar, bass, keyboard, and drums.
• The D1 Hand: A general-purpose five-finger dexterous hand with higher degrees of freedom, focused on flexible motion, independent multi-joint control, coordinated multi-finger manipulation, and tactile sensing.
• The M1 Hand: A modular dexterous hand designed for application-specific environments, using a standardized hand body plus interchangeable functional fingertips that can be configured for different operating tasks.

The company also previewed its F Series, which explores higher degrees of freedom, biomimetic structures, and compliant manipulation using a tendon-driven architecture designed to more closely replicate the movement characteristics of the human hand.

CASBOT Hands was established in July 2026 as a spinoff from CASBOT, a robotics company founded in 2023 that has spent several years developing embodied intelligence technologies and deploying them in real-world environments. The new company's strategy is to move dexterous hands beyond standalone hardware components and develop them into an integrated hardware-and-software manipulation platform, summarized as moving "from a pair of dexterous hands to a complete embodied manipulation capability".

Building a Complete Manipulation System, Not Just Hardware

CASBOT Hands describes its development model as a "multi-platform hardware plus data acquisition" dual-engine approach. On the hardware side, dexterous hands with different mechanical structures, degrees of freedom, and application targets provide a foundation for complete robots, robotics research, data collection, and real-world deployment. On the data side, tools such as data gloves, teleoperation systems, and data acquisition platforms are used to continuously collect manipulation data, supporting the iterative development of models, robotic skills, and hardware.

This approach reflects a broader industry realization: the future of embodied AI depends on moving beyond individual hardware specifications toward a comprehensive assessment of whether a robot can actually progress from grasping to manipulation to completing real tasks. The sector still faces challenges involving cost, performance, reliability, and the accumulation of high-quality manipulation data, but the capital investment and market demand suggest these obstacles are being addressed at scale.

Meanwhile, other parts of the robotics ecosystem are also advancing. Pudu Robotics, ranked number one globally in commercial cleaning robotics by revenue with a 29% market share, launched the PUDU ET1, an AI-native compact scrubber-dryer robot designed for small commercial spaces like convenience stores, drugstores, and chain restaurants. The robot integrates scrubbing, sweeping, vacuuming, and dust-mopping in one unit and features an automated 8-in-1 docking station that handles charging, water refilling, wastewater drainage, detergent dispensing, and self-cleaning cycles.

At the research level, the University of Hong Kong launched RoboDojo, a unified benchmarking platform designed to evaluate robotic manipulation across both simulated and physical environments. The platform encompasses 42 simulation tasks, 18 real-world robotic tasks, and 30 representative robot policies, assessing capabilities such as generalization, memory, precision, and long-horizon task execution. Initial findings highlight a significant performance gap between current robotic systems and human capabilities. The top-performing AI model achieved success rates of only 8.80% in simulation and 12.8% in real-world testing, compared to 76.03% and 100% respectively achieved by human experts.

"To the best of our knowledge, RoboDojo is the first Hong Kong-led benchmark to unify simulation and standardised real-robot evaluation. It moves embodied AI beyond impressive demonstrations towards progress that can be measured, compared and trusted," remarked Professor Ping Luo, Associate Director of AI Research and Tech Transfer at the HKU School of Computing and Data Science.

Professor Ping Luo, Associate Director (AI Research and Tech Transfer), HKU School of Computing and Data Science

The robotics industry is at an inflection point. The focus is shifting from building more complex hardware to building systems that can reliably perform real tasks. Dexterous hands are central to this transition, but only when they're part of a complete system that includes perception, force control, and accumulated task experience. The next wave of robotics progress will be measured not by specifications, but by what robots can actually do.