Imagine stepping into a world where robots handle everyday tasks with ease. Wang Xingxing, the founder of China’s Unitree Robotics, shared his vision during the World Robot Conference in Beijing, drawing comparisons to how ChatGPT shifted the AI landscape. He envisions a near future—maybe within the next year or two, possibly stretching to three—where robots could autonomously tidy up rooms or deliver items, even in unfamiliar settings.
While current robotics hardware and data are impressively advanced, the real challenge now is blending in the kind of ‘embodied intelligence’ that makes a machine truly independent. Wang is sceptical of today’s vision language action (VLA) models, describing their architecture as a bit “dumb” despite enhancements from reinforcement learning. One promising approach might involve creating interactive, video-driven models from text prompts to guide a robot’s movement more naturally.
Adding to the buzz, Wang pointed to Google’s recent launch of its Genie 3 ‘world model’—a tool capable of simulating dynamic environments with authentic physical properties. This step could signal a subtle yet impactful shift in robotics, making our interactions with technology smoother and more intuitive.
