AGIBOTWORLD2026 Physical Reasoning Dataset Release
Overview
Far征A3 Ultra, a full-sized bipedal humanoid robot, has been deployed in a commercial retail setting in Wuhan。
From Chinese robotics — Hardware and sensing
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · decision_justification_chain: 1 other · Basis not reported
“更绝的是，每条数据都附带‘决策依据链’：比如‘因检测到当前仅1盒且保质期剩余＜3天，故触发替换逻辑而非补货逻辑’。”
Exact source · revision 2Source owner not reported
Reported quantity · difficulty_level: 3 other · Basis not reported
“整套数据集覆盖九大核心推理能力：属性选择（选‘未开封’而非‘所有’酸奶）、条件判断（温度＞26℃则启动冷藏补偿）、操作可行性评估（抽屉卡滞→改用语音呼叫人工协助）、目标搜索（多模态线索融合定位）、长程规划（跨3个房间完成‘取药→烧水→送杯’闭环）、过程记忆（记住已开过的3个柜门）、状态维护（持续追踪‘已清洁’‘待消毒’‘已失效’等标签）……并按Easy（单条件分支）、Medium（双约束叠加）、Hard（含隐含规则+历史依赖）三级标定难度，堪称具身智能界的‘GRE逻辑推理题库’。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 74.51 percent · Basis not reported · Differing source assertions
“二、仿真与真机：双线领先 仿真评测：RoboTwin 2.0 Clean2Random 曾登顶榜单 ViGAR 在 RoboTwin 2.0 仿真评测的 Clean2Random 基准上取得 Clean 82.00%、Random 67.02% 的成功率，平均成功率 74.51%，超越最强基线 12.81 个百分点，发布时曾位列榜单第一；该榜单实时更新，截至发稿（2026 年 10 月）位列总榜第三。”
Exact source · revision 2Source owner not reported
“二、仿真与真机：双线领先 仿真评测：RoboTwin 2.0 Clean2Random 曾登顶榜单 ViGAR 在 RoboTwin 2.0 仿真评测的 Clean2Random 基准上取得 Clean 82.00%、Random 67.02% 的成功率，平均成功率 74.51%，超越最强基线 12.81 个百分点，发布时曾位列榜单第一；该榜单实时更新，截至发稿（2026 年 10 月）位列总榜第三。”
Exact source · revision 2Source owner not reported
Reported quantity · task_completion_progress: 0.69 other · Basis not reported
“在 AgiBot A2 全尺寸人形机器人上，ViGAR 对五个长时序组合任务进行真机评测，平均任务完成进度达到 0.69，领先 π0.5（0.60）和使用相同数据配方，但无子目标规划的 Cosmos3-Nano-Policy（0.54）。”
Exact source · revision 2Source owner not reported
“在 AgiBot A2 全尺寸人形机器人上，ViGAR 对五个长时序组合任务进行真机评测，平均任务完成进度达到 0.69，领先 π0.5（0.60）和使用相同数据配方，但无子目标规划的 Cosmos3-Nano-Policy（0.54）。”
Exact source · revision 2Source owner not reported
Reported quantity · training_hours: 900 hours · Basis not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
Reported quantity · tasks: 752 other · Basis not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
Reported quantity · objects: 200 other · Basis not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
Reported quantity · backgrounds: 10 other · Basis not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
“真机实验采用两条 7 自由度机械臂、两台腕部相机和一台前置相机；研究团队使用约 900 小时的同平台遥操作数据进行中期训练，覆盖 752 项任务、200 余种物体和 10 余种背景，再通过任务示范开展后训练。”
Exact source · revision 2Source owner not reported
Reported quantity · number of multimodal sequences: 120000 other · Basis not reported
“智元这次开源的AGIBOTWORLD2026-PhysicalReasoning，包含超12万段高清多模态序列（RGB-D+IMU+关节扭矩+语音指令+动作标注），覆盖37类真实扰动场景（宠物窜出、灯光骤暗、地面油渍、物品堆叠坍塌……），甚至加入了‘人类故意干扰测试’——比如突然抽走机器人正要抓取的托盘。”
Exact source · revision 2Source owner not reported
Reported quantity · number of real-world disturbance scenarios: 37 other · Basis not reported
“智元这次开源的AGIBOTWORLD2026-PhysicalReasoning，包含超12万段高清多模态序列（RGB-D+IMU+关节扭矩+语音指令+动作标注），覆盖37类真实扰动场景（宠物窜出、灯光骤暗、地面油渍、物品堆叠坍塌……），甚至加入了‘人类故意干扰测试’——比如突然抽走机器人正要抓取的托盘。”
Exact source · revision 2Source owner not reported
Reported quantity · annual service hours per unit: 4800 hours · Basis not reported
“特斯拉Optimus Gen-3已在加州三座汽车工厂执行巡检与物料搬运；宇树科技H1系列则拿下国内首个省级电力巡检全场景商用订单，单台年服务时长突破4800小时。”
Exact source · revision 2Source owner not reported
Reported quantity · zero bias stability of domestic MEMS inertial arrays: 0.003 other · Basis not reported
“答案藏在‘国产替代2.0’里：不是简单替换进口零件，而是用自研谐波减速器（传动效率提升至94.7%）、一体式空心杯伺服执行器（响应延迟<8ms）、国产MEMS六轴惯导阵列（零偏稳定性达0.003°/h）重构整机架构。”
Exact source · revision 2Source owner not reported
Reported quantity · response delay of domestic servo actuators: 8 other · Basis not reported
“答案藏在‘国产替代2.0’里：不是简单替换进口零件，而是用自研谐波减速器（传动效率提升至94.7%）、一体式空心杯伺服执行器（响应延迟<8ms）、国产MEMS六轴惯导阵列（零偏稳定性达0.003°/h）重构整机架构。”
Exact source · revision 2Source owner not reported
Reported quantity · orders completed: 827 other · Basis not reported
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 97 percent · Basis not reported · Differing source assertions
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · daily orders during peak: 100 other · Basis not reported
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 100 percent · Basis not reported · Differing source assertions
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · operational hours per day: 8 hours · Basis not reported
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · operational days: 18 other · Basis not reported
“（图片来源：智元） 记者获悉，截至10月19日，远征A3 Ultra已稳定运行18天，每日提供8小时服务精准匹配江滩夜游客流高峰；累计完成订单827单，整体抓取成功率超97%，周末客流高峰期间实现单日订单破百、抓取成功率100%。”
Exact source · revision 2Source owner not reported
Reported quantity · maximum operational hours: 8 hours · Basis not reported
“同时，依托UWB+RTK厘米级室内外融合定位、8小时商用续航，搭配自主充电、四重主动安全防护体系，设备可长时间在景区无人值守作业，大幅降低现场运维压力。”
Exact source · revision 2Source owner not reported
Reported quantity · dataset release: 4 other · Basis not reported
“AGIBOT Open-Sources AGIBOT WORLD Release 4: "Physical Reasoning" Dataset Enabling Robots to Reason Before Acting”
Exact source · revision 2agibot
Reported quantity · dataset categories: 10 other · Basis not reported
“ol, then determines which items are needed and directs a robot to pick and place them. A sunny forecast calls for sunscreen, while rain calls for an umbrella. Release 4 connects goal comprehension, task-state tracking, and action execution in real-world tasks. Covering over 10 categories of logical reasoning scenarios, the dataset deliberately includes multiple real-world states and execution branches for the same ta”
Exact source · revision 2agibot
Reported quantity · reasoning capabilities: 9 other · Basis not reported
“asp the key directly. If the key is blocked, the robot must first move the box. The goal remains the same, but the action path changes with the scene. Retrieving a key requires different action sequences depending on available space and obstructions. The dataset addresses nine core reasoning abilities: attribute-constrained selection, context-conditioned selection, manipulation feasibility assessment, action precondi”
Exact source · revision 2agibot
Reported quantity · task difficulty levels: 3 other · Basis not reported
“state maintenance, while finding an item involves search strategy, search history, and spatial conditions. Tasks are organized into three difficulty levels—Easy, Medium, and Hard. Easy tasks test individual judgments, such as attribute-constrained selection. Medium tasks introduce quantity requirements and action preconditions. Hard tasks combine multi-subgoal planning, long-horizon task-state tracking, and multiple”
Exact source · revision 2agibot
Artifact availability · dataset: available
“sk. By providing high-level Q&A and low-level action trajectories, it supports coordinated training of high-level decision-making models and low-level action policies, linking task decision-making to physical robot execution. The aim is to help robots both "reason clearly" and "execute successfully." Project Page: https://agibot-world.com Open Source Link: https://huggingface.co/datasets/agibot-world/AgiBotWorld2026”
Exact source · revision 2agibot
Evidence dependency changes (15)
- 2026-10-30T21:44:36.677Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T21:44:26.190Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:42:30.678Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:42:21.060Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:37:24.852Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:37:05.702Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:30:55.170Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:30:33.607Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:30:01.448Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:29:36.601Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:27:22.515Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:27:02.949Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:06:46.829Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:06:33.875Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-23T18:06:14.268Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
Developments
- 2026-10-22 09:36 UTC · 4 reportsAGIBOTWORLD2026 Physical Reasoning Dataset ReleaseAgiBot — Official news：AGIBOT Open-Sources AGIB T WORLD Release 4: 'Physical Reasoning' Dataset Enabling Robots to Reason Before Acting
- 2026-10-22 01:33 UTC · 2 reportsViGAR: Visual Goal-conditioned Action Reasoning for Full-Size Humanoid RobotsChinese robotics — Research news discovery：ViGAR: The First World-Action Model for Full-Sized Humanoid Robots
- 2026-10-20 21:23 UTC · 1 reportsAgibot Zhengyuan A3 Ultra Smart Retail Warehouse Deployment in WuhanChinese robotics — Hardware and sensing：Far征A3 Ultra Full-Sized Bipedal Humanoid Robot Deploys in Commercial Retail Scenario
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Dataset and collection discoveryZhiyuan Opens Source Embodied Reasoning Dataset. Humanoid Intelligence Enters New Era of 'Physical AI'
Zhiyuan Robotics officially opens sourced the fourth phase of AGIBOTWORLD2026 'Embodied Reasoning' dataset, which is the first global dataset covering multiple scenarios such as real homes, factories, and warehouses for training Physical Agents. It contains over 120,000 high-definition multimodal sequences, supporting commercial use and secondary annotation, providing the foundational capability for robots to 'think with their bodies.'
- AgiBot — Official newsSignalAGIBOT Open-Sources AGIB T WORLD Release 4: 'Physical Reasoning' Dataset Enabling Robots to Reason Before Acting
AGIBOT has released the fourth version of its AGIBOT WORLD dataset, focusing on 'Physical Reasoning' to enable robots to make decisions before acting. The dataset covers over 10 categories of logical reasoning scenarios, supporting both high-level decision-making models and low-level action policies. It includes annotations linking task goals to execution processes, aiding in the training of embodied agents for real-world tasks.
- Chinese robotics — Research news discoverySignalViGAR: First World-Action Model for Full-Scale Humanoid Robots
ViGAR, developed by Zhiyuan and Peking University, is the first complete world-action model (WAM) for full-scale humanoid bipedal robots. It has been validated on the real-world AgiBot A2 platform and achieved strong performance on the RoboTwin 2.0 benchmark. The model introduces a hierarchical architecture with subgoal planning and execution, enabling long-horizon task completion through explicit reasoning and context learning.
- Chinese robotics — Dataset and collection discoveryLet Robots Learn to 'Act Based on Circumstances': Zhiyuan Opens Fourth 'Physical Reasoning' Dataset
Zhuyuan Robotics has released the fourth 'Physical Reasoning' dataset in the AGIBOT WORLD 2026 series, focusing on decision-making under varying conditions. The dataset includes nine core reasoning capabilities and three difficulty levels, enabling robots to understand tasks, plan actions, and adapt to different scenarios.
- Chinese robotics — Dataset and collection discoverySignalLet Robots Learn to 'Act According to Circumstances'. Zhiyuan Opens AGIBOTWORLD2026 Fourth Physical Reasoning Dataset
On October 22, 2026, Zhiyuan officially released the AGIBOTWORLD2026 fourth 'Physical Reasoning' dataset, focusing on teaching robots to make decisions based on situational awareness. The dataset includes structured data with task phases, subtask boundaries, and decision rationale chains, covering nine core reasoning abilities and three difficulty levels. It supports advanced reasoning capabilities in robotics and is part of a series leading to the release of a new embodied AI framework.
- Chinese robotics — Research news discoverySignalViGAR: The First World-Action Model for Full-Sized Humanoid Robots
In October 2026, Zhuyuan Robotics collaborated with the DA Group at Peking University and CocoMatrix to release ViGAR, the first complete world-action model (WAM) for full-sized humanoid bipedal robots. ViGAR has been validated on the real-world AgiBot A2 platform and achieved strong results on the RoboTwin 2.0 benchmark, ranking third in the real-time leaderboard. The model introduces a hierarchical architecture with a subgoal planner and executor, enabling long-horizon task execution through explicit visual subgoal reasoning. This approach allows the robot to plan and execute complex, multi-step tasks in dynamic environments.
- Chinese robotics — Hardware and sensingSignalFar征A3 Ultra Full-Sized Bipedal Humanoid Robot Deploys in Commercial Retail Scenario
Far征A3 Ultra, a full-sized bipedal humanoid robot, has been deployed in a commercial retail setting in Wuhan。
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).