October 2026 Robotics Briefing: From Demonstrations to Dependable Work | 2026年10月机器人观察

22 OCTOBER 2026 · INDUSTRY SNAPSHOT

The next frontier is dependable work.

Physical AI is advancing. For households and businesses, the useful question is how a system performs across complete tasks, unfamiliar environments and months of service. This October review looks at three recent primary-source developments.

a humanoid and tactile-hand research in a robotics laboratory.

01 · Generalization is improving

On 17 September, Figure reported that Helix 2.5 completed three types of household tasks in 30 unseen homes with 56% success, compared with 9% without Index pretraining. These are Figure’s reported test results. The remaining failures matter as much as the improvement.

Our interpretation: a household pilot should log complete-task success, interventions and recovery—not only show a successful clip. Read Figure’s research ↗

02 · Data quality becomes infrastructure

Figure introduced Index on 25 August as a pipeline for human-demonstration data. Diverse examples may help robots transfer skills, but provenance, participant consent and evaluation quality remain important.

Our interpretation: design data governance alongside learning systems. Read the Index announcement ↗

03 · Plan for the entire robot lifecycle

On 30 September, Figure announced retiring its F.02 fleet as its newer platform expanded. Rapid hardware change makes maintenance, spare parts, replacement and end-of-life terms practical buying questions.

Our interpretation: compare service coverage and lifecycle cost alongside capability. Read the lifecycle announcement ↗

Watch the original demonstrations

Figure logistics demonstration, linked by Figure ↗ · Boston Dynamics Atlas playlist ↗

External demonstrations and research belong to their publishers. They are not Neurobatic performance claims or partnership announcements.

2026年10月22日 · 行业观察

下一阶段,是可靠地完成工作。

具身智能持续进步。对家庭与企业而言,关键是系统能否在完整任务、陌生环境与长期服务中稳定运行。本期回顾三项近期原始来源进展。

01 · 泛化能力进步

Figure于9月17日报告,Helix 2.5在30个未见家庭中执行三类任务,完整任务成功率为56%,未进行Index预训练的对照组为9%。这是Figure公布的测试结果。失败情况与进步幅度同样值得关注。

我们的判断:家庭试点应记录完整任务成功率、人工干预与恢复过程,而非仅展示成功片段。

02 · 数据质量成为基础

Figure于8月25日介绍人类示范数据计划Index。多样样本有助于技能迁移,同时须关注数据来源、参与者同意与评估质量。我们的判断:数据治理应与学习系统同步设计。

03 · 规划完整生命周期

Figure于9月30日宣布退役F.02机队。快速迭代让维护、备件、更换及报废条款成为实际采购问题。我们的判断:评估能力时,也应比较服务覆盖与全生命周期成本。

上方链接提供原始来源与官方视频。第三方演示不代表灵动天工的产品能力或合作关系。

What this means for Neurobatic · 对灵动天工的意义

Our 2028–2030 service roadmap starts with defined tasks, measurable pilots and human oversight. Explore the right deployment path for your home, business or factory.

2028–2030年服务路线,从明确任务、可衡量试点与人工监督开始。探索适合家庭、企业与工厂的部署路径。