Family World Simulator

Family world simulator is Yueban Dongli’s simulated home-interaction environment for training and testing Xiaoban before large amounts of real household data exist. 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】, Shibo says the simulator models active and passive emotional interactions, predicts changes in human emotional state and interaction state, and generates data used to fine-tune the fast and slow brains.

Role In The Product

• Provides synthetic or simulated interaction data for Emotional Interaction Models.
• Helps test long-term companion behavior before wide household deployment.
• Extends World Models from physical reasoning into family relationship state, emotion, and interaction dynamics.
• Reduces dependence on collecting raw household audio or images, though the episode leaves open how robust the privacy and data pipeline will be in practice.

Connections

• Xiaoban and Yueban Dongli — product and company case.
• On Device Fast Slow Brain — model stack tuned with simulator data.
• Companion Robots and Robot Liveliness — user-facing goals the simulator supports.
• Embodied AI and World Models — broader technical context.