A dedicated facility for robotic education, described as a "kindergarten for robots," has officially opened its doors on September 1st in Beijing's Shougang Park, Shijingshan District. Rather than simply grouping many robots together for standard training, this innovative center aims to establish a mechanism that allows machines to learn continuously through perception, action, feedback, and reflection—much like a human child. A firsthand visit to the facility reveals a structured environment designed to nurture autonomous robotic development.
The facility is organized into three distinct functional zones: a testing area, a learning area, and an interaction area. The bright and lively color scheme inside closely resembles that of a human kindergarten, creating a familiar and welcoming atmosphere. Currently, fifteen robots from various brands, including Unitree, Accelerated Evolution, and JAKA, are adapting to their new surroundings, with most featuring compact designs such as quadruped robotic dogs and small humanoid forms. Some of the diminutive humanoids, wearing bibs, appear particularly endearing as they pace back and forth, wave to visitors, or playfully immerse themselves in a ball pit.
HOU Guangdong, vice president of R&D at Tashan Technology, explained that the approach is not about mimicking human movements. Instead, robots are encouraged to understand themselves, their environment, and other agents, allowing them to acquire skills through their own cycles of sensing, acting, and reflecting, just as children do. Before a robot can engage in genuine autonomous learning, it must address the most fundamental issue: ensuring its physical body can support extended periods of independent exploration. Because no off-the-shelf solutions exist for this requirement, the primary task begins with upgrading the robot hardware platform to make it more durable, or "tougher."
The testing area serves as the place to build this hardware foundation. The goal is to create a robust robotic platform capable of running continuously for long periods, upon which algorithm development can proceed. This allows robots to ceaselessly collect environmental data, make autonomous decisions, and learn without interruption. Hardware modifications focus on two main tasks: reinforcing the body with a flexible outer shell to improve drop resistance, and installing tactile skin. While fifteen robots are currently enrolled, plans indicate that over thirty more will enter the facility by the end of the year for similar hardware enhancements.
A particularly notable robot on-site wears a sleeve decorated with a puppy pattern, providing a tangible demonstration of the "tactile skin" technology. When a gentle touch is applied, the robot responds with a cheerful greeting, and when its arm is squeezed more firmly, it verbally reacts as if feeling discomfort. HOU explained that the tactile skin, developed independently by Tashan Technology, serves as a vital sensory organ for exploring the outside world and is a critical component of the robot's safety protocols. For autonomous exploration, touch is as essential to robots as it is to young children, as it is the earliest sensory input humans acquire.
This tactile sense, which gathers information on pressure, temperature, and humidity, complements visual data. All robots entering the program are equipped with this electronic skin, enabling them to react appropriately to different levels of external contact—a crucial step toward meaningful environmental interaction. HOU noted that most robots are currently in the testing and hardware renovation phase. Once they possess comprehensive multimodal perception and a stable, long-running hardware platform, they can advance to the "learning area," a safe interactive space designed for independent exploration and skill acquisition.
The methodology here diverges sharply from conventional robotic training. HOU pointed out that the common industry practice involves offline collection of human-demonstrated data to train a model, which is then deployed for specific applications. However, once deployed, the robot's intelligence becomes static and frozen, unable to adapt to the ever-changing realities of the physical world or its own component wear and tear. The true objective is to enable online, continuous learning so robots can perpetually adjust to environmental shifts and changes in their own physical condition.
After mastering new skills, robots advance to the most challenging zone: the interaction area. Here, they are expected to learn from each other, mirroring how children observe parents and mimic peers. The future vision includes robots capable of collaborating and observing each other to complete tasks, a high-level expression of general intelligence. This capability is seen as essential for adapting to complex settings like family homes, where robots will need to self-adjust to environmental changes without human instruction, relying on mutual observation, prediction, and cooperation to solve problems without direct commands.
HOU believes that touch is not just a sensory endpoint but the origin of control for embodied intelligence. To integrate touch into a robot's ability to generalize, it is necessary to build a complete technical system—spanning chips, sensors, data, algorithms, and training paradigms—dedicated entirely to the sense of touch. In the field of tactile perception, Tashan Technology's monthly delivery volume of tactile sensors has reached tens of thousands, with orders in the first half of the year exceeding four times the total for the entire previous year and a market share above 80 percent.
Despite its market position, the company does not define itself merely as a sensor vendor. It has progressed to the stage of turning touch into a universal, generalized capability. HOU revealed that the team has developed the world's first dynamic tactile sensing chip and is currently tackling technological breakthroughs in real-time continuous learning algorithms, continuous environmental modeling, and autonomous goal setting. This "robot kindergarten" model provides a complete training environment from tactile perception to skill acquisition, enabling robots to autonomously accumulate operational experience through real physical interactions and reducing reliance on human-demonstrated data.