The Zhitong Finance App learned that Guosheng Securities released a research report saying that it is recommended to focus on hand-eye and brain integration, dexterous hands, tactile perception, and micromotor modules. The essence of the physical interactive brain is to fill the control vacuum between the general model and the underlying physical execution, and push the robot from “being able to demonstrate” to “capable work.” The advent of hand-eye and brain integration technology has once again emphasized the investment value of contact-level sensing and control in the industrial chain. It is recommended to focus on: 1) hand-eye brain integration; 2) tactile perception; 3) dexterous hands & micromotor modules related targets.
Guosheng Securities's main views are as follows:
The GPT-6 Astra marks that the general model has crossed the threshold of physical operation, but there is still a cliff in “last centimeter” fine contact control. There is an urgent need for a new technical paradigm that is independent of the big model and uses tactile sensation as the native input
Astra has been able to directly generate motion commands to control robotic arms, and has achieved a generational leap forward in spatial reasoning tasks such as grabbing blocks, and the success rate far exceeds that of previous generations. However, in the face of tasks requiring millimeter-level contact adjustment, such as inserting puzzles, the success rate plummeted to 10%. This shows that the existing general model has a huge gap in physical control after contact, and simple spatial reasoning cannot replace contact-level control. As a “last centimeter” physical carrier, the core pain points of dexterous hands are tactile feedback and real-time force control. Therefore, the future breaking point of the industry is no longer simply expanding the parameters of the VLA model, but moving towards constructing a new paradigm of physical interaction brain with tactile perception as the core input that can truly solve contact-level control problems.
Dexterous hands are not simple hardware engineering, but intelligent carriers of robot physical interaction
The multi-finger multi-joint mechanical structure and high degree of freedom only give the robot basic physical movement space, which is a necessary but not sufficient condition to achieve complex tasks. In a real unstructured environment, facing the sliding, deformation, or position shift of an object, it is impossible to complete general precision assembly and handheld operation by simply relying on a preset trajectory or mechanical movement with a high degree of freedom. This requires dexterous hands to break through pure actuator positioning and have the ability to sense contact conditions in real time, understand the environment, and correct underlying dynamic feedback. As a result, the industrial logic of dexterous hands is spanning from simple mechanical performance competition to comprehensive intelligent control integrating “perception, decision, and execution”. Its essence has evolved into a software and hardware collaborative system with deep integration of multi-modal sensing and real-time algorithms.
The “closed loop of physical interaction” constructed collaboratively by hand, eye and brain is the core of breaking through the bottleneck of refined robot operation, and Daimeng's robot's “physical interaction brain” provides key technical verification for this path
Facing high-touch tasks such as precise insertion and flexible body operation, single-point capabilities that simply rely on vision, large models, or dexterous hardware have reached the boundary, and the industry urgently needs to achieve deep integration of “perception - decision - execution - feedback”. As a typical example of this paradigm, Daimon-TWM (Tactile Anchored World Model) released by Daimon Robotics in August 2026 integrates physical perception, deductive decisions, and instantaneous control into a unified framework to truly create a “physical interactive brain” that drives the robot's dexterous operation. Test data showed that the model achieved an average success rate of 64% and 53% in undisturbed and disturbed environments for contact-intensive tasks, verifying that predictive decision-making and high-frequency closed-loop control significantly improved the operational fault tolerance rate. This marks a complete shift in physical AI competition from competition on a single hardware parameter or model scale to system-level collaboration barriers centered on hand-eye and brain integration.
Risk warning: robot application model development falls short of expectations; risk over time; risk of sample error.