Robotics Bottleneck Research机器人瓶颈研究

Main主页 / companies / candidates / dexmal-yuanli-lingji

candidate

Dexmal原力灵机

The Dexbotic VLA toolkit and DOS-W1 robot hardware, both open-sourced, and co-operation of RoboChallenge, a large-scale real-robot evaluation benchmark, with Hugging Face.

开源的 Dexbotic VLA 工具链与 DOS-W1 机器人本体,并与 Hugging Face 共同运营大规模真机评测基准 RoboChallenge。

  • PROVISIONAL — NOT VERIFIED待定条目 — 未经核验

This record is provisional.本条目为待定状态。 It hasn't been through the checks the rest of the map has, so it carries no source references or diligence score, and the index won't hold it. The blockers below say what would have to change. 它还没走完地图其余部分都走过的核验流程,所以没有来源引用,也没有尽调评分, 指数不会持有它。下面的阻塞项写着要先解决什么。

What it doesThe Dexbotic VLA toolkit and DOS-W1 robot hardware, both open-sourced, and co-operation of RoboChallenge, a large-scale real-robot evaluation benchmark, with Hugging Face.

在干什么开源的 Dexbotic VLA 工具链与 DOS-W1 机器人本体,并与 Hugging Face 共同运营大规模真机评测基准 RoboChallenge。

Why it mattersP16, evaluation, was scored near-empty in the file, and RoboChallenge is the largest real-robot evaluation venue that exists: the Table30 suite has powered more than 80,000 real-robot task executions across 20-plus models on a public leaderboard, explicitly testing cross-robot generalisation. Evaluation infrastructure is where a field's claims get priced, and the file currently has no node for it.

为何重要P16(评测)在本文件中几乎是空的,而 RoboChallenge 是现存规模最大的真机评测场:Table30 套件已支撑 20 余个模型、超过 8 万次真机任务执行,榜单公开,并明确测试跨本体泛化。评测基础设施是一个领域的主张接受定价之处,而本文件目前没有对应节点。

Caveat

It is vendor-operated evaluation. Dexmal also ships models and hardware, so leaderboard governance is a live conflict of interest, the same pattern the file already flags for Lightwheel/RoboFinals and Daimon/RobOmni.

A benchmark run by a participant is a marketing surface until an independent party holds the scoring. Founded March 2025, so there is no operating history to diligence.

Close to RMB 1bn was raised across A and A+ rounds announced 2025-11-14, with NIO Capital leading the A and Alibaba the A+; no valuation was disclosed and none should be inferred.

风险提示

这是供应商自办的评测。原力灵机同时出模型与硬件,榜单治理因此存在现实的利益冲突——本文件已对光轮智能的 RoboFinals 与戴盟的 RobOmni 标注过同一模式。

由参赛方运营的基准,在评分权交给独立第三方之前,本质上是营销界面。公司 2025 年 3 月成立,没有可供尽调的经营历史。

2025-11-14 公布 A 与 A+ 两轮合计近 10 亿元融资,蔚来资本领投 A 轮,阿里巴巴领投 A+ 轮;估值未披露,也不应据以推算。

Bottlenecks瓶颈 P06 Robot data collection机器人数据采集 · P11 Generalisable VLA policies可泛化的 VLA 策略 · P16 Safety validation & certification安全验证与认证

Proposed for
companies
Class
private
Country
CN
HQ
Beijing, China
Ticker confidence
confirmed
Systems
blood · brain
Evidence
[object Object] · [object Object] · [object Object]
Proposed by
brain-senses-blood
Confidence
85
Translated fields
what_it_does_zh · why_it_matters_zh · caveat_zh