Yushu Technology CEO: Robots can generate any action and autonomously initiate attacks, achievable in six months

"The most worthwhile thing for us to promote right now, which can also yield quick results, is optimizing robot movements. In about six months, we will be able to achieve arbitrary movement generation for robots." Recently, at the Siemens RXD conference held in Beijing, Wang Xingxing, founder, chairman, and CEO of Yushu Technology, stated this during a conversation with Siemens global CEO Roland Busch. Wang Xingxing explained that currently, when Yushu robots compete in combat matches, their movements are quite mechanical and fixed, with all actions pre-collected. "For example, we have collected over twenty movements, and during battles, the robot relies on these movements to combine attacks, but the moves are relatively fixed. The same punching method is used every time, which lacks challenge and visual appeal." He mentioned that in the current optimization process, they will first collect hundreds of movements for AI training. Once training is complete, these movements will allow for free and smooth combinations. The robots will be able to flexibly perform various coherent actions such as punching, changing direction, and dodging up, down, left, and right, significantly enhancing their agility. At that time, the combat moves of the robots will be different every day, and the richness of the actions between two fighting robots will be much higher, making it extremely visually appealing. "Everyone should feel the same way; if a robot can only perform a few fixed actions, its level of intelligence is very low. However, if a robot can perform hundreds, thousands, or even millions of different actions, and can freely combine and make autonomous decisions on attacks, then its level of intelligence is very high, which is the core key." Wang Xingxing emphasized that the richness of movements directly determines the level of intelligence of the robot. This is crucial for humanoid robots to enter factory and everyday application scenarios more quickly. "I firmly believe that only when robots can perform a variety of rich movements, combined with AI technologies such as large language models, allowing the system to call these combined movements, can robots truly execute practical tasks." He stated. Wang Xingxing, founder of Yushu Technology, converses with Siemens global CEO Roland Busch The conference showcased the application of humanoid robots in industrial settings: based on Siemens SIMOVE Fleet Manager intelligent scheduling platform, it enables unified scheduling and collaborative management of automated guided vehicles (AGVs) and Yushu humanoid robots on factory sites. This solution explores key capabilities such as path planning and obstacle avoidance, communication, and coordination after the robots enter the factory, presenting a pathway for embodied intelligence in industrial scenarios. Wang Xingxing stated that the issues of humanoid robot movement and basic actions have basically been conquered, but the technical challenges related to grasping and operation, especially those involving tactile feedback, have yet to be overcome. This is also a key bottleneck that restricts the large-scale deployment of humanoid robots in factories and homes. "Robots are not incapable of manipulating objects; for trained items, as long as the training is adequate, the success rate of grasping can reach nearly 100%. However, once there is a slight change in the object, the success rate will drop sharply. To solve this problem, a massive amount of training data on different objects is needed to fill the gaps, but currently, this part of the data training has not been fully conquered." He stated that the walking, running, and various kung fu movements of the humanoid robot are primarily trained in a purely simulated environment. However, for practical tasks, such as having the robot grasp objects or assemble components, the current global simulation technology is still not mature enough, and most still rely on real human data collection for training. The challenge with real data collection is significant, as the number of realistic scenarios that can be constructed is limited, making it impossible to replicate thousands of scenarios in reality, and the costs of construction and data collection are too high. "I believe both technical routes (simulation and real scene training) are worth advancing; currently, the industry has not formed a unified optimal solution. Tactile simulation technology is also crucial, as it can more accurately simulate the object grasping process, which is a core focus area at present." Xiao Song, President and CEO of Siemens Greater China, told The Paper that humanoid robots may achieve simple factory tasks (such as handling, screwing, polishing, etc.) within one or two years, but to be highly reliable and deeply integrated into the core production system, undertaking complex and precise work will still require 5-10 years of technological accumulation and ecological co-construction. "This requires their technical performance to meet the stringent demands of industrial standards: nearly 100% success rate, precision and efficiency far exceeding current levels, and absolute safety and reliability. At the same time, they need to adapt to various industrial scenarios, not just solve a few preset problems. This relies on the maturity of training data, simulation environments, and the entire ecosystem," he said to The Paper. Risk Warning and Disclaimer The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at one's own risk