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】

Summary

This Shizilukou Crossing road-podcast episode has Koji interviewing Shibo, founder of Yueban Dongli, about Xiaoban, a consumer bipedal Companion Robots product designed as a family partner rather than a household tool, pet replacement, or novelty toy. The discussion frames Robot Liveliness as the core product problem: language, movement, soft materials, eyes, charging, memory, privacy, and personality must all make the robot feel like an independent individual. It also gives a concrete technical architecture for companion robotics through Qwen, On Device Fast Slow Brain, Emotional Interaction Models, and a Family World Simulator.

Key Claims

• Companion Robots should not be evaluated only as task robots; Xiaoban is intentionally designed around companionship, emotional response, and household coexistence rather than manipulation or domestic labor.
• Shibo treats excessive human-like language as a liability for companion products, so Xiaoban uses a 12-sound non-human language paired with gaze, posture, belly-tapping, and other embodied cues.
• The product’s short arms, soft body, bipedal movement, expressive eyes, and sitting charger all reinforce Robot Liveliness by making the robot feel safe, touchable, and less like a configurable appliance.
• Xiaoban is built to avoid simple command-and-entertainment patterns; it should initiate interaction, react emotionally, remember people, and develop different personality tendencies over time.
• The architecture combines an on-device 1.7B “fast brain” for immediate behavior decisions with a 7B “slow brain” for more complex reasoning, aiming for voice, movement, and bodily feedback latency under 0.4 seconds.
• Yueban Dongli uses Qwen because Shibo says the open-source model performs well for small-parameter reasoning and emotion understanding, then adapts it with home-scene interaction data.
• The team uses a Family World Simulator to model long-term active and passive emotional interactions before there is enough real household data from shipped devices.
• Shibo argues that companionship is not the same as pleasing the user; if a child repeatedly mistreats Xiaoban, the robot may become less willing to initiate interaction with that child.
• As of the episode, the domestic price was 8499 RMB, the preorder price discussed was 6499 RMB, and the fastest planned delivery window was around China’s National Day holiday on October 1, 2026.
• The episode’s unresolved risks are long-term retention, mass-production reliability, after-sales support, safety boundaries, and whether semantic-feature privacy protections are enough for an always-present household robot.

Key Quotes

“如果 Lavot 会说一个字或一句话，他可能就不会买。” — Shibo explaining why restraint can strengthen a companion robot.

“务实基础上干浪漫的事。” — Shibo on Yueban Dongli’s team culture.

“No excuse for engineer.” — the engineering standard Shibo says shaped him from university.

“谁不想让哆啦A梦走到现实里。” — Shibo on the emotional pull behind building Xiaoban.

Connections

• Koji — host of the road-podcast conversation.
• Shibo — guest and founder explaining the product, technical architecture, and company philosophy.
• Yueban Dongli — company building Xiaoban.
• Xiaoban — central product case for consumer Companion Robots and Robot Liveliness.
• LOVOT — reference companion robot whose simple interaction logic influenced Shibo’s thinking about restraint.
• Marc Raibert and Boston Dynamics — robotics inspiration for Shibo.
• Qwen and Open Source AI Models — model base used for the small-parameter emotional reasoning stack.
• Embodied AI, World Models, and Voice Interaction — existing wiki themes extended by the episode into household companion robotics.
• On Device Fast Slow Brain, Emotional Interaction Models, and Family World Simulator — technical concepts surfaced by the episode.
• Product Led Willingness To Pay — the host’s “first truly want-to-buy companion robot” reaction and the product’s sub-10000 RMB pricing make this a consumer hardware willingness-to-pay case.

Contradictions

• None identified. The episode extends prior Embodied AI and World Models themes from investment watchlists into a specific consumer companion robot case rather than contradicting them.