Figure 机器人训练流程曝光,演示能力被认为超过 Optimus
有人晒了一家人形机器人公司的演示,觉得比 Tesla 的 Optimus 还能打。真正有意思的是他们的训练闭环:戴手套采人类技能,仿真里找模型崩溃点再加固。想搞具身智能数据管线的人可以看看这套思路。
Santiago 在 X 上分享了一组人形机器人演示,认为其能力已经超过 Tesla 的 Optimus。帖子只提到演示观感,没有给出参数量、基准名称或发布日期等具体数字。
文中介绍了这家公司训练机器人的四步工作流:工作人员戴手套采集人类操作技能,用这些数据训练模型,再通过仿真找出模型失效的场景,最后把失效反馈用于加固模型。这条"采集—训练—仿真压测—反馈强化"的闭环是该流程的核心。
这个说法来自一条社交媒体帖子,属于个人观感而非第三方评测,两个机器人也没有可比的公开基准成绩。对关注具身智能数据采集方式的开发者来说,用手套捕获人类技能加上仿真找边界的做法值得留意。
These guys have one of the most capable robots I've seen. Based on the demos, Optimus is nowhere near it.
They have a pretty awesome workflow to teach these robots:
• People with gloves capture human skills • They use these skills to train a model • They use simulations to find where the model breaks • They use the feedback to harden the model
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include humanoid robots, Optimus, Figure, simulation training.
Source attribution: curated from Ate-a-Pi (@svpino).