On Device Fast Slow Brain

On-device fast-slow brain is the architecture described by Shibo for Xiaoban in 我遇到了第一个真正想买的陪伴机器人！｜对话世博：越伴动力创始人【公路播客】. The fast brain is a 1.7B model for immediate behavior decisions, while the slow brain is a 7B model for more complex logic and reasoning.

Why It Matters

• Companion Robots need low-latency embodied response because slow reaction breaks the feeling of presence.
• The episode says Xiaoban can keep voice, movement, and body feedback latency under about 0.4 seconds.
• The architecture supports 5-to-10-second generated action sequences instead of repeating the same canned animation every time.
• It pairs with cloud long-term memory and Emotional Interaction Models, but the source says raw audio and image data are not collected; training uses semantic features and embeddings.

Connections

• Xiaoban and Yueban Dongli — product and company using the architecture.
• Qwen and Open Source AI Models — model ecosystem feeding the architecture.
• Robot Liveliness — product quality that depends on fast embodied response.
• Family World Simulator — simulated home data source used to tune behavior.