Key Info

Google's Gemma team shared that NVIDIA built on Gemma 4 E4B to put live conversation, tool calls, animations and a vision model onto a single 8GB Jetson device inside a robot.

Highlights

  • A Gemma-based model was placed in a robot body and immediately scanned the room
  • NVIDIA's build runs on Gemma 4 E4B
  • Live conversation, tool calls, animations and a vision model all fit on one 8GB Jetson
  • Demonstrates capable multimodal inference on low-power edge hardware