OpenMind and Tashan Technology open ‘robot kindergarten’ in Beijing, leaderobot.com reports

leaderobot.com reported on September 14, 2026, that a ‘robot kindergarten’ had opened in Shougang Park in Shijingshan, Beijing. It said the facility was launched by Tashan Technology and OpenMind, founded by Richard Sutton, the 2024 Turing Award winner, and featured robots from Unitree Robotics and Booster Robotics, among others.

If robot training shifts from a focus on human imitation toward iterative learning in real-world environments, what is needed is not just the robot itself but a learning infrastructure that includes sensing, safety and continuous operation. The framework reported by leaderobot.com suggests that competition in embodied intelligence could extend beyond model accuracy to the learning environment itself.

What is new in this report is its presentation of an embodied intelligence training hub not as a place to collect imitation data, but as a permanent learning space designed for trial and error in real-world environments. At the same time, the idea of placing robots in a safe space to test their behavior is an extension of an established approach, so the story appears to emphasize how research is organized rather than represent a wholly new development. See the original report for details. In related coverage, we reported on humanoid.guide’s teardown of the Unitree G1, focusing on actuators and cooling design and humanoid.guide’s report on BeyondMimic for the Unitree G1, which showed how attention has shifted from finished-product presentation to internal design and training methods. The ‘robot kindergarten’ takes that trend further into hands-on learning environments, suggesting that the focus is less on showcasing robot performance than on how to design repeated cycles of sensing, action and feedback. The implications for the industry are that competition could extend beyond robot specifications to include learning environments, sensing modules such as tactile systems, and safety design that can support long operating hours. Tashan Technology’s use of tactile skin suggests that sensory systems on the input side can serve as a starting point for learning, but whether they can be extended into a genuinely general-purpose learning mechanism is a separate question. Unresolved questions include how learning outcomes at the facility will be defined in terms of specific capabilities, how broadly they can be transferred across robot platforms, and what evaluation metrics will be used under conditions close to real-world operation. The handling of training data, safety standards and the range of participating robots also appear to require confirmation from the original report.

For Japanese companies and research institutions, developing robot hardware may not be enough; tactile sensors, safety design and experimental environments that support long operating hours could become relatively more important. In particular, designing a learning space that minimizes the risk of damaging physical robots also affects preproduction evaluation and data collection methods.