Interview with Luo Ping: The Core of Embodied AI Models is Bridging the Human-to-Robot Data Loop

机器之心 · wechat · 2026-08-03

Jiqizhixin conducted an in-depth interview with Luo Ping, co-founder of an embodied AI startup, exploring their full-stack approach to large embodied models. At WAIC, the company's full-size humanoid robot, KAIBot, demonstrated zero-error can-opening, relying on a training pipeline where 90% of the experience came from first-person human data.

Luo positions the company as an "embodied large model company," aiming to build a model and training platform adaptable to various robot morphologies rather than just selling hardware. To minimize information loss during human-to-robot skill transfer, KAIBot features a highly anthropomorphic design with 117 degrees of freedom and extensive tactile skin, enabling safe physical interactions and robust feedback collection.

Before leveraging Scaling Laws, the company focuses on bridging the human-robot data gap. They deploy hundreds of data collectors in real-world environments to gather first-person data, which is then processed via an automated pipeline for VLA and world model training. They also introduced KAIWorldModel, combining 4D spatial modeling with video generation to extend data production into virtual worlds for rare scenarios. This full-stack strategy tightly couples hardware, data, and algorithms to accelerate the robot evolution cycle.

Original post →

Interview with Luo Ping: The Core of Embodied AI Models is Bridging the Human-to-Robot Data Loop

More from Embodied