Blog

After eight months of building a robot dog, I walked two shows with one question: June's Computex and August's Taiwan Robotics and Intelligent Automation Show in Nangang. There were plenty of robots, but fewer actually moving than I expected; quadrupeds multiplied, yet almost all of them were for inspection; and many booths ran an architecture nearly identical to my own PawAI.

I spent eight months building an interactive robot dog on a Unitree Go2 with a Jetson and a stack of sensors. This post uses measured data to lay out every difficulty across hardware, sensors, navigation, voice and testing: three power converters, a memory wall that turned out, once measured, to be a CPU limit, a blind spot in every sensor, and why integrating so many capabilities still left us without a concrete use case.

In early 2025 I was still following React tutorials; by the end of the year I was using code to make a 30 kg robot dog stand up in my living room. That year I landed my first internship, cut a sign-up flow from 7 steps to 5, wrote my first backend, bought my first 3D printer, and picked a robot dog for my capstone. Getting stronger didn't happen all at once; it accumulated step by step, and only shows when you look back.