Guojin Securities: I am optimistic about digital gloves and dexterous hands in the medium term high barriers+high value aspects of physical AI

The Zhitong Finance App learned that Guojin Securities released a research report saying that the integration of VLA (vision-language-motion model) and the world model is expected to push robots from single-step operation to multi-step planning, failure recognition, and autonomous recovery. Digital sampling gloves and dexterous hands are not an alternative relationship; they are two complementary digital acquisition methods, “human side” and “machine side”. I am optimistic that the robot's tactile requirements will drive the volume of digital acquisition devices such as digital collection gloves and dexterous hands. Driven by the development of physical AI, digital gloves and dexterous hands are expected to become core incremental links in industry trends.

Guojin Securities's main views are as follows:

GPT-6 Astra makes up for the shortcomings of the robot's “brain”. The next stop will be System 1 evolution driven by physical interaction data

The GPT-6 Astra succeeded in 49 of the 50 episodes, with a success rate of 98%, while the single pi0.5 model had a success rate of only 36%, proving that the general-purpose large model acts as a robot System 2 (slow thinking brain), and the “brain doesn't work” problem that has limited the industry for a long time has reached an inflection point. Physical AI, represented by the world model, expands the AI capability chain from “perception-understanding” to “perception-prediction-planning-control”. By integrating VLA (vision-language-action model) with the world model, it is expected to push robots from single-step operation to multi-step planning, failure recognition, and autonomous recovery. Industry consensus is gradually converging into a hybrid architecture of “general reasoning and dedicated operation” — the big model is the top-level perception of System 2, the physical model is the physical execution of System 1, and if System 1 is to evolve, it must be fed with physical interaction data that cannot be simulated.

Digital gloves and dexterous hands are expected to become core incremental links in physical AI

The core requirements of physical AI are physical state data — data such as contact force, tangential force, deformation, contact timing, etc. These tactile data require actual interactive collection and labeling. At the beginning of 2026, the global stock of high-quality real physical interaction data was about 500,000 hours, while general physical model training required at least 10 million hours, which is less than 5% of the demand, and there is a high demand for tactile acquisition related equipment in the medium to long term. The core devices for tactile data collection include digital collection gloves (human-side collection) and dexterous hand remote control (aircraft-side collection). Among them, digital gloves are hand-worn, and the cost is lower than that of dexterous hands. They can enter various actual scenarios through manual data collection, mainly pre-training and tactile characterization and learning; dexterous mobile phone robot tactile data sources. The core advantage is accurate motion tags based on the robot body, no retargeting, and biased towards fine strategy tuning and real machine verification. Digital sampling gloves and dexterous hands are not an alternative relationship; they are two complementary digital acquisition methods, “human side” and “machine side”. I am optimistic that the robot's tactile requirements will drive the volume of digital acquisition devices such as digital collection gloves and dexterous hands.

Investment advice and valuation

Driven by the development of physical AI, digital gloves and dexterous hands are expected to become core incremental links in industry trends. Relevant targets include Cixing Co., Ltd., Dari Electronics, Fulai New Materials, and Changying Precision. (1) Cixing Co., Ltd.: Computer knitting machine leader, which relies on three-dimensional knitting technology to extend to flexible sensing fabrics. The company acquired Swiss Steiger in 2010 and signed a contract to acquire assets related to the German STOLL brand in June 2026 to supplement high-end knitting equipment, 3D weaving and fabric design capabilities, and expand application fields such as smart wear and flexible sensing. (2) Dari Electronics: A supplier of consumer electronics structures and functional components, with precision die-cutting, flexible material processing and multi-material composite manufacturing capabilities. Related technology is common with the flexible substrates, sensor fixing and structural protective materials required for tactile digital gloves. (3) Fulai New Materials: Relying on polymer materials and precision coating technology to lay out flexible tactile sensors, the products have been delivered in batches. In February 2026, a procurement cooperation was reached with Smart Hand for 100,000 sets of tactile sensors. By the end of August, more than 30,000 sets had been delivered, mainly supporting industrial-grade dexterous hands and expanding to tactile data collection terminals. (4) Changying Precision: Relying on precision processing and multi-material molding capabilities, the products cover humanoid robot dexterous hand transmission parts, joint core transmission parts, and trunk skeletal structural parts. At the same time, it has expanded to force-control joint modules. Related products have completed multi-wheel sampling and testing, involving torque sensing, force feedback, and execution control.

Risk Alerts

The development of the robotics industry falls short of anticipated risks; progress in the promotion of large models falls short of expectations; progress in research and development of digital gloves and dexterous hands falls short of expectations.