Skild AI：为机器人打造通用大脑

Skild AI: Building a Universal Brain for Robots

本期速览 · Overview

Skild AI 创始人探讨其使命：为任何机器人打造通用大脑，将机器人技术视为数据问题。

Skild AI founders discuss their mission to create a general-purpose brain for any robot, treating robotics as a data problem.

机器人学是数据问题；通用大脑可驱动任何机器人，但部署是主要挑战。
Robotics is a data problem; a universal brain can power any robot, but deployment is the main challenge.
在多样化数据上预训练，再在真实世界数据上后训练，是机器人学习的关键配方。
Pretraining on diverse data and post-training on real-world data is the key recipe for robot learning.
与语言模型不同，机器人部署不能像软件那样扩展；它需要严格的测试和时间。
Unlike language models, robotics deployment cannot scale like software; it requires rigorous testing and time.

机器人技术是一个数据问题；没有机器人数据的互联网，所以我们必须利用所有可用的数据来源。
Robotics is a data problem; there is no internet of robot data, so we must use all available data sources.
一个通用大脑，就像语言领域的ChatGPT，可以为任何形态的机器人和任务提供动力。
A general-purpose brain, like ChatGPT for language, can power any robot across form factors and tasks.
部署是机器人技术的主要挑战；它需要严格的测试，不能像软件那样快速扩展。
Deployment is the primary challenge in robotics; it requires rigorous testing and cannot be scaled like software.
在多样化数据（视频、模拟）上进行预训练，并在真实世界数据上进行后训练是关键方法。
Pretraining on diverse data (videos, simulation) and post-training on real-world data is the key recipe.
跨垂直领域的数据飞轮使得从结构化到非结构化环境的扩展成为可能。
A data flywheel across verticals enables scaling from structured to unstructured environments.

与语言模型不同，机器人技术不能一夜成功；部署需要时间，是一个技术挑战。
Unlike language models, robotics cannot achieve overnight success; deployment takes time and is a technical challenge.
仅靠视频数据是不够的；我们需要模拟和真实世界数据来弥合差距。
Video data alone is insufficient; we need simulation and real-world data to bridge the gap.
家用机器人的中期时间线高度不确定；即使是专家也意见不一。
The middle-term timeline for home robots is highly uncertain; even experts disagree.
在机器人技术中，部署本身就是一个技术挑战，而不仅仅是商业问题。
Deployment itself is a technical challenge, not just a business problem, in robotics.
人类在短期内乐观，长期悲观，但机器人技术的进展甚至让专家感到惊讶。
Humans are optimistic in the short term and pessimistic in the long term, but robotics progress surprises even experts.