Taking Undergrad Students to Google's Global AI Competition: We Lost, Of Course, But I'm Truly Impressed!
Earlier I led two undergrad placement students to compete in Google’s global AI competition. We lost in the end, of course, but it was an excellent hands-on experience for the two students! The most impressive part was seeing the winning Japanese team’s project, which once again showed just how formidable Japan’s depth is in combining hardware and software (Physical AI / Embodied AI).
🏆 Winning project: GEM-4 (Gemma Embodied 4 Physical Assistance)
They applied Google’s latest Gemma model to Physical Assistance (physical assistance / robot control). It’s not just chatting in front of a screen: they turned the LLM into the robot’s “brain” to understand real-world physical environments, handle multimodal instructions, and precisely control the robot to help with daily life.
💡Technical takeaways: 1️⃣ Embodied AI is the next big trend: Software AI is no longer limited to text and images. Getting AI to “touch the ground” and interact with the real world is the real Game Changer. 2️⃣ The Japanese team’s craftsmanship: Details were handled with extreme rigour, from Prompting and Multimodal data processing to cutting control-loop latency, and the polish was so high it felt like a commercial-grade product. 3️⃣ The biggest value of taking students along: Even though we lost, letting them experience first-hand the gap against top international teams is far more useful than saying “the world out there is big” a hundred times in a classroom.
Failure is the mother of success. Next time, we fight again! 💪🔥
📄 Winning writeup link (a must-read if you’re into Robotics / Multimodal AI): https://lnkd.in/gPWJRnxf