Axis中文俱乐部由社区驱动， 致力推动 Axis Robotics在华语区的发展， 引入华语力量，提升Axis Robotics全球影响力。@axisrobotics
Axis Robotics
Joined December 2024
- Tweets439
- Following9
- Followers400
- Likes259
我们正在构建一个多层次的机器人架构，其中大语言模型驱动高级推理和技能利用。
这是一个演示。任务：使用右臂将托盘中的三个物体沿桌子的 X 轴排列成一行，按从最高到最矮的高度排序。
你可以看到代理通过技能利用将任务转化为连续运动，将场景理解、基于高度的排序以及放置规划与物理执行相连接。
这是一个令人兴奋的开始。 @axisrobotics
We're building a multi-layered robotics architecture, where the LLM drives high-level reasoning and skill harness.
Here's a demo. The task: use the right arm to arrange three objects from the tray in a row along the X axis of the table, ordered by height from tallest to shortest.
You can see the agent turn the task into continuous motion through skill harness, linking scene understanding, height-based ordering, and placement planning to physical execution.
This is an exciting start.
. base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 的账户已上线 @HTX_Global ，可获得最高 150%的年利率 ⬇️
宣布 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 代币回购
我们的数据引擎正在为现实世界场景中的物理 AI 应用提供动力。我们希望每位贡献者都能直接从中受益。
我们正在启动 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 代币回购计划，第一批 15 万美元的回购资金已完成。
1. 我们的商业交易采用先试点后推广的模式。只有在试点成功后才会签署框架协议，并且付款将根据交付里程碑分批释放。
2. 从今以后，我们商业合同的一部分净利将在收到现金时用于回购 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea。
每季度将发布一份公开的回购报告，链上资金记录可由任何人独立验证。所有回购的 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 代币将被转移至公共资金库。
资金库地址：0xfC8c6ABC0742E64c13d6B0bAC1fCD7e604eC7Dc5
查看完整详情：docs.axisrobotics.ai/economi…
@axisrobotics
Announcing the $AXIS Token Buyback
Our data engine is powering Physical AI applications across real-world scenarios. We want every contributor to share directly in it.
We're launching the $AXIS token buyback, with our first $150K tranche already complete.
1. Our commercial deals follow a pilot-first model. Framework agreements are signed only after a successful pilot, and payments are released in tranches against delivery milestones.
2. Going forward, a portion of the net profit from our commercial contracts is allocated to buy back $AXIS as cash is collected.
A public buyback report is published each quarter, with on-chain treasury records anyone can independently verify. All repurchased $AXIS is transferred to the public treasury.
Treasury address: 0xfC8c6ABC0742E64c13d6B0bAC1fCD7e604eC7Dc5
See the full details: docs.axisrobotics.ai/economi…
由我们的顶级贡献者 ⬇️ 撰写的完整教程。⬇️
Some Tips and Tricks I use on @axisrobotics
i tried to simplify the process as much as possible😎
产品更新 🛠️
我们减少了开放任务槽的数量。原因如下。
1️⃣ GPT-6 现在可以解决简单任务并为其生成可靠的轨迹。
2⃣ 高质量的人类演示对于复杂任务仍然至关重要。这些演示的价值远超常规轨迹。
第一季专注于多样化的人类输入用于预训练。在这个基础确立后，我们的重点转向强化学习、专家和蒸馏。更严格的槽位结构直接支持这一转变，并将资源集中在高质量反馈和实际价值上。
查看我们的高层技术路线图 ⬇️ @axisrobotics
Product Update 🛠️
We've reduced the number of open task slots. Here's why.
1️⃣ GPT-6 can now solve simple tasks and generate reliable trajectories for them.
2️⃣ High-quality human demonstrations remain essential for complex tasks. These demonstrations carry far more value than routine trajectories.
Season 1 focused on diverse human input for pretraining. With that foundation in place, our focus moves to RL, experts, and distillation. A tighter slot structure supports this shift directly, and keeps resources concentrated on high-quality feedback and practical value.
See our high-level technical roadmap ⬇️
There's a new version of this post
VLAs 如何从自然次优的机器人演示中学习？
介绍 AXIS-Bench：质量感知 VLA 预训练中什么最重要，这是一个基准测试和实证研究，探讨演示质量应如何定义、标注，并在预训练中使用。
在三个 VLAs 和 24 个下游任务中，我们系统比较了质量感知学习选择，并发现没有一种单一的配方适用于每个模型。
与来自 @GeorgiaTech @CarnegieMellon @UCBerkeley
@NTUsg @NUSingapore @RiceUniversity @UCRiverside @TAMU 的研究人员合作。
详情 & 配套数据集发布 ⬇️
How should VLAs learn from naturally suboptimal robot demonstrations?
Introducing AXIS-Bench: What Matters in Quality-Aware VLA Pretraining, a benchmark and empirical study of how demonstration quality should be defined, labeled, and used during pretraining.
Across three VLAs and 24 downstream tasks, we systematically compare quality-aware learning choices and find that no single recipe works for every model.
A collaboration with researchers from @GeorgiaTech @CarnegieMellon @UCBerkeley @NTUsg @NUSingapore @RiceUniversity @UCRiverside @TAMU.
Details & Companion Dataset Release ⬇️
6/7 数据发布
Axis 数据集 V1 于 8 月推出。此后，我们构建了数据集 V2，现覆盖 1,092 个任务和 2.35M 个回合（包括相机增强）。AXIS-Bench 中使用的配套数据集均从 V2 池中精心挑选：
• 预训练数据集。5,000 个仿真 Franka 演示，跨越 731 个任务，带有质量标签和帧索引。
• 基准数据集。354 个微调演示，跨越 24 个留出任务，带有任务定义和成功检查器、场景资产以及评估布局。
• 仿真到现实预训练数据集。19,250 个可重放模拟回合，跨越 663 个任务，随时可重新渲染到您自己的工作空间中，用于真实演示微调前的预训练。
这两项工作服务于互补目标：Axis 扩展数据资源，而 AXIS-Bench 研究数据质量和训练选择如何影响下游 VLA 性能。
7/7 链接
探索 AXIS-Bench 和我们更广泛的数据资源：
项目资源：
项目页面：axis-bench.github.io
论文：arxiv.org/abs/2610.18869
代码：github.com/axis-bench/axis-b…
试用 Axis Hub：hub.axisrobotics.ai
数据集：
Axis 数据集 V2：huggingface.co/datasets/axis…
预训练：huggingface.co/datasets/axis…
基准：huggingface.co/datasets/axis…
仿真到现实预训练：huggingface.co/datasets/axis…
我们希望 AXIS-Bench 能促进更多关于从不完美演示中学习的、可重复的研究。
@axisrobotics
关于10月28–29日 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 抛售的声明
过去两天，一系列地址在链上出售了大量 base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea。我们希望清楚地说明这些代币的来源以及我们正在采取的措施。
1. 没有出售团队或投资者代币。 所有核心团队和投资者的分配，以及所有其他锁定的分配，总计835,000,000 AXIS（占总供应量的83.5%），仍保留在与Binance共同管理的归属合约中：0xAFE41cfd4C21f1D896678e32C5AD456ea41835e6。任何人都可以在BaseScan上验证余额。没有一枚代币离开该合约，也没有代币在12个月的悬崖期之前解锁。
2. 出售代币的来源。 出售的代币来自我们公布的代币经济学中规定的初始流通供应量，具体来说是生态系统分配的一部分。许多生态系统接收者并非加密原生用户，因此他们的分配是通过单一OTC交易对手根据书面协议交付的，该协议包括锁定期条款。
3. 我们认为发生了什么。 销售模式——通过交易所存款路径，并在交付后数小时内涉及数百个新地址——强烈表明OTC交易对手在公开市场上出售了这些代币。如果是这样，这与协议不符。
4. 我们正在做什么。 - 我们已暂停与该交易对手的所有进一步交易。 - 我们正在与法律顾问审查协议，并将追究协议下可用的每一项补救措施。 - 今后，任何剩余的生态系统代币将仅通过链上归属进行分配。 团队的代币仍处于锁定状态，我们构建Axis的承诺不变。
Statement on the $AXIS Sell-off of 28–29 October
Over the past two days, a series of addresses sold a large volume of $AXIS on-chain. We want to set out clearly what these tokens were and what we are doing about it.
1. No team or investor tokens were sold.
All Core Team and investor allocations, together with every other locked allocation, 835,000,000 $AXIS (83.5% of total supply), remain in the vesting contract, which is jointly administered with Binance: 0xAFE41cfd4C21f1D896678e32C5AD456ea41835e6. Anyone can verify the balance on BaseScan. Not a single token has left it, and none unlocks before the 12-month cliff.
2. Where the sold tokens came from.
The tokens sold came from the initial circulating supply set out in our published tokenomics, specifically part of the Ecosystem allocation. Many ecosystem recipients are not crypto-native, so their allocation was delivered through a single OTC counterparty, under a written agreement that includes lock-up terms.
3. What we believe happened.
The pattern of the sales, which went through exchange deposit routes and hundreds of fresh addresses within hours of delivery, strongly suggests that the OTC counterparty sold these tokens on the open market. If so, this is inconsistent with the agreement.
4. What we are doing.
- We have suspended all further dealings with this counterparty.
- We are reviewing the agreement with legal counsel and will pursue every remedy available under it.
- Going forward, any remaining Ecosystem tokens will be distributed only through on-chain vesting.
The team‘s tokens remain locked, and our commitment to building Axis is unchanged.
Axis对机器人进行训练，维罗纳验证数据真实有效。
Axis机器人公司 🤝 Verona @axisrobotics
base:0x5be75e3b0aff5d1a042a5363ae36acb1d6d755ea 交易竞赛现已在 #Binance Alpha 开启。
尽早参与活动，最高可获得 3.0 倍的积分奖励。
Hub @huggingface 现已支持 OpenUSD 格式，
@nvidiaomniverse SimReady 资产已在上面上线。
这清楚表明，在物理 AI 中，数据是核心瓶颈，而模拟数据的价值正日益被认可和挖掘。
Axis 更进一步：我们正在构建连接仿真与现实的数据引擎，将仿真轨迹一直带入现实世界的部署。 @axisrobotics
The @huggingface Hub now supports OpenUSD format, with @nvidiaomniverse SimReady assets already live on it.
This is a clear sign that data is the core bottleneck in Physical AI, and that the value of sim data is increasingly being recognized and tapped.
Axis takes it a step further: we're building the data engine that bridges sim and real, carrying simulation trajectories all the way into real-world deployment.