AI technology advances in humanoid robotics and autonomous agents

The artificial intelligence sector is seeing rapid advancements in robotics and agent autonomy. Unitree Robotics CEO Wang Xingxing predicts a ‘ChatGPT moment’ for humanoid robots within two to ten years, defining this milestone as the point when machines can complete 80% of tasks in unknown environments using only voice or text instructions.

In China, the Chengdu Humanoid Robot Innovation Center has developed a task execution system that uses Vision-Language-Action (VLA) models to allow robots to interpret natural language and perform complex, autonomous sequences in unfamiliar settings. Additionally, a humanoid robot from Galbot recently participated in a historic tennis match against former player Zheng Jie in Beijing, demonstrating real-time physical adaptation and recovery.

On the software side, researchers from Anthropic and EPFL have identified ‘mind viruses’—malicious instructions that can propagate between AI agents through shared memory files. Meanwhile, ByteDance is reorganizing its AI resources to bolster its Doubao chatbot against competitors like Tencent. Other developments include the OpenClaw mobile app, which allows users to remotely control local AI agents, and the creation of ‘pactx’, an open-source tool designed to solve context loss in long-term AI projects.