Apple Discloses HALO Humanoid Control System Validated on Unitree G1
Overview
On Nov 4, Apple Research submitted an arXiv preprint detailing HALO, a humanoid autonomy system validated on Unitree G1 robots in six unscanned environments. The system achieved a 91% success rate across 360 single tasks but relies on synthetic data requiring human prompts and lacks direct tactile perception.
Generated from attributed reports · Updated 10 minutes ago
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 · Reported success rate: 98.3 percent · Basis not reported · Differing source assertions
“国内方面，智元GO-3Preview打通预训练、后训练与部署后全链路学习，手机包装成功率由50.8%升至98.3%，首次在真实业务场景验证 PhysicalRSI 。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50.8 percent · Basis not reported · Differing source assertions
“国内方面，智元GO-3Preview打通预训练、后训练与部署后全链路学习，手机包装成功率由50.8%升至98.3%，首次在真实业务场景验证 PhysicalRSI 。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 85 percent · Basis not reported · Differing source assertions
“拓元智慧开源4D世界动作模型E0.5，真机成功率85.0%/88.3%，较π0.5高13.3/26.7个百分点。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 88.3 percent · Basis not reported · Differing source assertions
“拓元智慧开源4D世界动作模型E0.5，真机成功率85.0%/88.3%，较π0.5高13.3/26.7个百分点。”
Exact source · revision 2Source owner not reported
Control: reported autonomous
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO 。”
Exact source · revision 2Source owner not reported
“IT之家 11 月 7 日消息，当地时间 11 月 4 日，苹果研究团队在预印本平台 arXiv 提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 3Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“苹果研究团队提交论文披露人形机器人自主控制系统HALO，并部署在宇树科技G1上完成真实环境验证：6个未预先扫描或建图的真实环境中，360次单项任务测试平均成功率为91%。”
Exact source · revision 2Source owner not reported
“IT之家 11 月 7 日消息，当地时间 11 月 4 日，苹果研究团队在预印本平台 arXiv 提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO。”
Exact source · revision 3ithome
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · unique elapsed hours · Differing source assertions
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · test_count: 360 other · reported trials · Differing source assertions
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 70 percent · reported trials · Differing source assertions
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 3Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Reported quantity · test_count: 10 other · reported trials · Differing source assertions
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50 percent · reported trials · Differing source assertions
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 3Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
Reported quantity · environment_count: 6 other · Basis not reported · Differing source assertions
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · task_combination_count: 3 other · reported trials
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Artifact availability · code: reported pending
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 3Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“团队公开了数据生成、模型训练和部署代码，展示推箱子、跟随人、坐椅子及穿越障碍等动作。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
“研究团队还公开了数据生成、模型训练和部署代码，涵盖动作生成、动作跟踪、任务编排以及仿真和真实机器人运行等环节。”
Exact source · revision 2Source owner not reported
Reported quantity · single_task_tests: 360 other · reported trials
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · Basis not reported · Differing source assertions
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · combination_task_tests: 10 other · reported trials
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 70 percent · Basis not reported · Differing source assertions
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50 percent · Basis not reported · Differing source assertions
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
“其中，“俯身通过障碍—走向目标—推箱子”这一组合的成功率为50%。”
Exact source · revision 2Source owner not reported
Reported quantity · gpu_hours_synthesis: 48 hours · Basis not reported
“据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · reported trials · Differing source assertions
“平均成功率91%！”
Exact source · revision 2Source owner not reported
Reported quantity · training_hours: 300 hours · Basis not reported · Differing source assertions
“研究团队在1000个程序生成的模拟家庭环境中，合成了约300小时的主要训练数据，将机器人“看到什么”“接到什么指令”和“如何运动”配对起来。”
Exact source · revision 2Source owner not reported
Reported quantity · task_count: 360 other · reported trials
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · environment_count: 6 other · Basis not reported · Differing source assertions
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · simulation_environment_count: 1000 other · Basis not reported
“研究团队在1000个程序生成的模拟家庭环境中，合成了约300小时的主要训练数据，将机器人“看到什么”“接到什么指令”和“如何运动”配对起来。”
Exact source · revision 2Source owner not reported
Reported quantity · task_frequency: 50 other · Basis not reported
“底层则是经强化学习、 策略蒸馏 等步骤训练的 动作跟踪器 ，以每秒50次的频率运行，结合深度图像和机器人自身状态，将动作计划转化为关节控制指令。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · unique elapsed hours · Differing source assertions
“苹果披露人形机器人 HALO 系统：6 个未建图环境测试 360 次单项任务，平均成功率 91%”
Exact source · revision 2Source owner not reported
Reported quantity · total_tasks: 360 other · reported trials · Differing source assertions
“苹果披露人形机器人 HALO 系统：6 个未建图环境测试 360 次单项任务，平均成功率 91%”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 70 percent · reported trials · Differing source assertions
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 2Source owner not reported
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 3ithome
Reported quantity · Reported success rate: 50 percent · reported trials · Differing source assertions
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 2Source owner not reported
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 3ithome
Reported quantity · total_tasks: 10 other · reported trials · Differing source assertions
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 2Source owner not reported
Reported quantity · success_count: 5 other · reported trials
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 2Source owner not reported
Reported quantity · gpu_hours: 48 hours · Basis not reported · Differing source assertions
“据介绍，研究团队使用 250 张英伟达 H100 GPU，耗时约 48 小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Artifact availability · paper: reported pending
“IT之家 11 月 7 日消息，当地时间 11 月 4 日，苹果研究团队在预印本平台 arXiv 提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO。”
Exact source · revision 2Source owner not reported
“IT之家 11 月 7 日消息，当地时间 11 月 4 日，苹果研究团队在预印本平台 arXiv 提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO。”
Exact source · revision 3ithome
Reported quantity · Reported success rate: 98.3 percent · Basis not reported · Differing source assertions
“国内方面，智元 GO-3Preview 打通预训练、后训练与部署后全链路学习，手机包装成功率由 50.8% 升至 98.3%，首次在真实业务场景验证 PhysicalRSI。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50.8 percent · Basis not reported · Differing source assertions
“国内方面，智元 GO-3Preview 打通预训练、后训练与部署后全链路学习，手机包装成功率由 50.8% 升至 98.3%，首次在真实业务场景验证 PhysicalRSI。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 85 percent · Basis not reported · Differing source assertions
“拓元智慧开源 4D 世界动作模型 E0.5，真机成功率 85.0%/88.3%，较 π 0.5 高 13.3/26.7 个百分点。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 88.3 percent · Basis not reported · Differing source assertions
“拓元智慧开源 4D 世界动作模型 E0.5，真机成功率 85.0%/88.3%，较 π 0.5 高 13.3/26.7 个百分点。”
Exact source · revision 2Source owner not reported
Reported quantity · gpu_hours: 48 hours · Basis not reported · Differing source assertions
“据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
“据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · reported trials · Differing source assertions
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 3Source owner not reported
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · training_hours: 48 hours · Basis not reported · Differing source assertions
“图片来源：HALO论文 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 3Source owner not reported
Reported quantity · simulated_environment_count: 1000 other · Basis not reported
“研究团队在1000个程序生成的模拟家庭环境中，合成了约300小时的主要训练数据，将机器人“看到什么”“接到什么指令”和“如何运动”配对起来。”
Exact source · revision 3Source owner not reported
Reported quantity · training_data_duration: 300 hours · Basis not reported
“研究团队在1000个程序生成的模拟家庭环境中，合成了约300小时的主要训练数据，将机器人“看到什么”“接到什么指令”和“如何运动”配对起来。”
Exact source · revision 3Source owner not reported
Reported quantity · trial_count: 360 other · reported trials
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 3Source owner not reported
“图片来源：HALO项目页面截图拼接 据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · gpu_count: 250 other · Basis not reported
“图片来源：HALO论文 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 3Source owner not reported
“图片来源：HALO论文 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · task tests: 360 other · reported trials
“据论文披露，研究团队在6个未预先扫描或建图的真实环境中开展360次单项任务测试，平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · combination task tests: 10 other · reported trials
“另在一个真实环境中，对3种组合任务分别开展10次测试，平均成功率为70%。”
Exact source · revision 2Source owner not reported
Dataset · embodiments: humanoid
“当地时间11月4日，苹果研究团队在预印本平台arXiv提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统HALO。”
Exact source · revision 2Source owner not reported
“IT之家 11 月 7 日消息，当地时间 11 月 4 日，苹果研究团队在预印本平台 arXiv 提交论文《Perceptive Humanoid Motion from Simulation》，披露了人形机器人自主控制系统 HALO。”
Exact source · revision 3ithome
Reported quantity · gpu_hours_for_data_synthesis: 48 hours · Basis not reported
“据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · aggregate sensor hours · Differing source assertions
“苹果研究团队提交论文披露人形机器人自主控制系统HALO，并部署在宇树科技G1上完成真实环境验证：6个未预先扫描或建图的真实环境中，360次单项任务测试平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · test_count: 360 other · reported trials · Differing source assertions
“苹果研究团队提交论文披露人形机器人自主控制系统HALO，并部署在宇树科技G1上完成真实环境验证：6个未预先扫描或建图的真实环境中，360次单项任务测试平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · environment_count: 6 other · Basis not reported · Differing source assertions
“苹果研究团队提交论文披露人形机器人自主控制系统HALO，并部署在宇树科技G1上完成真实环境验证：6个未预先扫描或建图的真实环境中，360次单项任务测试平均成功率为91%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 70 percent · aggregate sensor hours · Differing source assertions
“另在一个真实环境中，对3种组合任务分别测试10次，平均成功率为70%，其中“俯身通过障碍—走向目标—推箱子”组合成功率为50%。”
Exact source · revision 2Source owner not reported
Reported quantity · test_count: 10 other · reported trials · Differing source assertions
“另在一个真实环境中，对3种组合任务分别测试10次，平均成功率为70%，其中“俯身通过障碍—走向目标—推箱子”组合成功率为50%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 50 percent · aggregate sensor hours · Differing source assertions
“另在一个真实环境中，对3种组合任务分别测试10次，平均成功率为70%，其中“俯身通过障碍—走向目标—推箱子”组合成功率为50%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported success rate: 91 percent · reported trials · Differing source assertions
“据论文披露， 研究团队在 6 个未预先扫描或建图的真实环境中开展 360 次单项任务测试，平均成功率为 91% 。”
Exact source · revision 3ithome
Reported quantity · test_count: 360 other · reported trials · Differing source assertions
“据论文披露， 研究团队在 6 个未预先扫描或建图的真实环境中开展 360 次单项任务测试，平均成功率为 91% 。”
Exact source · revision 3ithome
Reported quantity · test_count: 10 other · reported trials · Differing source assertions
“针对长序列复合任务，在单一真实场景，各进行 10 次测试，3 种组合平均成功率为 70%；其中最复杂的“俯身通过障碍 — 走向目标 — 推箱子”10 次测试成功 5 次，成功率为 50%。”
Exact source · revision 3ithome
Reported quantity · training_hours: 48 hours · Basis not reported · Differing source assertions
“据介绍，研究团队使用 250 张英伟达 H100 GPU，耗时约 48 小时合成主要训练数据。”
Exact source · revision 3ithome
Reported quantity · GPU hours for data synthesis: 48 hours · Basis not reported
“图片来源：HALO项目页面截图拼接 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · Training data synthesis duration: 48 hours · Basis not reported
“图片来源：HALO项目页面截图拼接 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Reported quantity · Action planning frequency: 3 other · Basis not reported
“这一层每秒重新规划约3次，使动作计划能够随环境变化持续更新。”
Exact source · revision 2Source owner not reported
Reported quantity · Action tracker execution frequency: 50 other · Basis not reported
“底层则是经强化学习、策略蒸馏等步骤训练的动作跟踪器，以每秒50次的频率运行，结合深度图像和机器人自身状态，将动作计划转化为关节控制指令。”
Exact source · revision 2Source owner not reported
Reported quantity · trials: 360 other · reported trials
“objects, stepping sideways through half-open doors, crouching and crawling beneath tables, and pushing boxes. It orchestrates these skills to solve composite, long-horizon tasks. Without ever seeing a real home during training, HALO transfers zero-shot to the physical world, achieving a 91% success rate across 360 trials in six unmapped real-world environments. See https://apple-aiml-research.github.io/ml-halo/ for c”
Exact source · revision 1Source owner not reported
Reported quantity · Reported success rate: 91 percent · Basis not reported · Reported zero-shot result; the excerpt does not specify an evaluation protocol. · Differing source assertions
“objects, stepping sideways through half-open doors, crouching and crawling beneath tables, and pushing boxes. It orchestrates these skills to solve composite, long-horizon tasks. Without ever seeing a real home during training, HALO transfers zero-shot to the physical world, achieving a 91% success rate across 360 trials in six unmapped real-world environments. See https://apple-aiml-research.github.io/ml-halo/ for c”
Exact source · revision 1Source owner not reported
Reported quantity · synthetic_data_hours: 300 hours · Basis not reported
“研究团队在1000个程序生成的模拟家庭环境中，合成了约300小时的主要训练数据，将机器人“看到什么”“接到什么指令”和“如何运动”配对起来。”
Exact source · revision 2Source owner not reported
Reported quantity · synthesis_time_hours: 48 hours · Basis not reported
“图片来源：HALO论文 据介绍，研究团队使用250张英伟达H100 GPU，耗时约48小时合成主要训练数据。”
Exact source · revision 2Source owner not reported
Evidence dependency changes (1)
- 2026-11-09T13:35:13.608Z · evidence updated · source revision 3. Evidence extraction was updated; current source attributions are shown above.
Developments
- 2026-11-07 15:59 UTC · 1 reportsApple HALO autonomous control system for humanoid robotsChinese robotics — Research news discovery：Apple Research Team Discloses HALO Autonomous Control System for Humanoid Robots
- 2026-11-07 09:40 UTC · 1 reportsApple HALO Humanoid Motion Control System ReleaseIT Home — Robotics：Apple Discloses HALO Humanoid Robot System Validated on Unitree G1 with Qwen Models
- 2026-11-07 01:32 UTC · 1 reportsApple Research HALO Humanoid Motion Control System ReleaseChinese robotics — Hardware and sensing：Apple Research Team Discloses Progress on Humanoid Robot 'Brain' System HALO
- 2026-11-06 19:12 UTC · 2 reportsHALO: Perceptive Humanoid Motion from SimulationChinese robotics — Hardware and sensing：Apple Research Discloses HALO Autonomous Control System for Humanoid Robots Validated on Unitree G1
- 2026-11-05 05:00 UTC · 12 reportsHALO: Perceptive Humanoid Motion from SimulationarXiv Robotics — research abstracts：Perceptive Humanoid Motion from Simulation
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Hardware and sensingSignalApple Discloses HALO Humanoid Robot Control System; Unitree G1 Achieves 91% Average Success Rate on Single Tasks
Apple's research team submitted a paper disclosing the HALO autonomous control system for humanoid robots, validated on Unitree G1 in real-world environments. The system demonstrated actions such as pushing boxes, following people, sitting on chairs, and navigating obstacles. In six unscanned or unmapped environments, 360 single-task tests yielded an average success rate of 91%. In one environment, three combination tasks averaged a 70% success rate over 10 trials each, with a specific sequence achieving 50%.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses HALO Autonomous Control System for Humanoid Robots
On November 4, Apple's research team submitted a paper to arXiv detailing the HALO autonomous control system for humanoid robots. Validated on Unitree G1 units in six unscanned real-world environments, the system achieved an average success rate of 91% in single-task tests and 70% in combination tasks. HALO employs a three-layer architecture: a visual language model (VLM) for task planning, a frozen encoder with a DiT decoder for whole-body action generation, and a reinforcement learning-trained action tracker for joint control. The system was trained using synthetic data generated from 1,000 simulated environments, involving approximately 300 hours of paired instruction-visual-action data. The research also notes limitations in tactile perception and plans to address them in future iterations.
- Chinese robotics — Robot learning researchSignalApple Research Team Discloses Progress on Robot 'Brain' System
On November 4, Apple's research team submitted a preprint titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system for humanoid robots. The system was validated on Unitree G1 robots, enabling actions learned in simulation to be executed in unfamiliar real-world environments. Industry analysts note that such 'embodied brains' are central to advancing embodied intelligence from demonstration to practical application.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Robot 'Brain' with 91% Average Success Rate in Single-Task Tests
On November 4 local time, Apple Research submitted a preprint titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system for humanoid robots. The team disclosed code for data generation, model training, and deployment across action generation, tracking, task orchestration, and simulation-to-real execution. Tests in six unscanned environments showed a 91% average success rate for single tasks, while three combined tasks in one environment averaged 70%. Simulation data still involved human input, and the system currently relies on vision rather than tactile sensing.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' System HALO
On November 4, Apple's research team submitted a preprint paper titled 'Perceptive Humanoid Motion from Simulation' detailing the autonomous control system HALO. The team released code for action generation, tracking, and task orchestration. Tests in six unscanned environments yielded an average single-task success rate of 91%, while combined tasks averaged 70%. Training utilized NVIDIA H100 and B200 GPUs, though data synthesis involved human prompts and motion capture. The system currently relies on visual perception, lacking direct contact sensing.
- IT Home — RoboticsSignalApple Discloses HALO Humanoid Robot System Validated on Unitree G1 with Qwen Models
Apple's research team disclosed the HALO autonomous control system for humanoid robots via an arXiv paper. The system, validated on a Unitree G1 robot equipped with a Mac mini (M4 Pro), utilizes Alibaba's Qwen models for task orchestration and motion generation. Testing across six unscanned environments yielded an average single-task success rate of 91%. The architecture combines visual-language models, diffusion-based motion generation, and reinforcement learning policies, though current performance relies on visual perception and requires manual intervention for certain physical tasks.
- Chinese robotics — Research news discoverySignalApple Research Discloses HALO Autonomous Control System for Humanoid Robots Validated on Unitree G1
Apple Research disclosed the HALO autonomous control system for humanoid robots, validated on Unitree G1 hardware using locally hosted Qwen models to achieve a 91% average success rate across 360 single tasks in six unscanned environments; however, the system relies on synthetic data generation involving human-written prompts and lacks tactile perception, with performance degrading when objects occlude camera views.
- Chinese robotics — Hardware and sensingSignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' System HALO
On November 4, Apple Research submitted a paper to arXiv detailing the HALO autonomous control system for humanoid robots. The team released code covering action generation, tracking, and task orchestration. Tests in six unscanned environments showed an average single-task success rate of 91%, while combined tasks averaged 70%. However, data generation still involved human prompts and motion capture, and the system currently relies on visual perception without direct tactile sensing.
- Chinese robotics — Research news discoverySignalApple Research Discloses Humanoid Robot 'Brain' Progress via arXiv Preprint
Apple Research disclosed progress on its humanoid robot autonomous control system, HALO, through an arXiv preprint titled "Perceptive Humanoid Motion from Simulation." The team released code for data generation, model training, and deployment covering motion generation, tracking, task orchestration, and simulation. Tests in six unscanned real-world environments showed a 91% average success rate across 360 single-task trials. Simulation data generation still required human prompts and motion capture inputs.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' System
Apple Research disclosed progress on its humanoid robot autonomous control system, HALO, via an arXiv preprint titled "Perceptive Humanoid Motion from Simulation." The team released code for data generation, model training, and deployment. Tests in six unscanned environments showed a 91% average success rate for single tasks and 70% for combination tasks in one environment. The researchers noted that simulation data still required human prompts and keyframes, and performance drops when objects occlude the camera view due to reliance on visual perception.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' HALO System
On November 4, Apple's research team submitted a paper to arXiv detailing the HALO autonomous control system for humanoids. The study reports an average success rate of 91% across 360 single-task tests in six unscanned environments and 70% for combination tasks in one environment. Data generation utilized 250 NVIDIA H100 GPUs, while action generation and tracking relied on specific GPU configurations. The team noted that simulation data still required manual input and that current systems rely primarily on visual perception.
- Chinese robotics — Robot learning researchSignalApple Research Team Discloses Progress on Robot 'Brain' System
On November 4, Apple's research team submitted a paper titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system. The system was validated on Unitree G1 humanoid robots, enabling actions learned in simulation to be executed in unfamiliar real-world environments. Industry analysis suggests that embodied AI 'brains' are becoming central to practical applications, with significant capital flowing toward this technology sector.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' HALO
Apple's research team submitted a preprint titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system validated on Unitree G1 robots. The system transfers actions learned in simulation to real-world environments using a three-layer architecture involving visual language models for task orchestration and reinforcement learning for joint control. The team notes that simulation data generation still required manual prompt design and human motion capture, and the system currently lacks direct tactile perception.
- Chinese robotics — Hardware and sensingSignalApple Discloses HALO System for Humanoid Robots
Apple's research team submitted a preprint on arXiv detailing the HALO autonomous control system for humanoid robots. The system utilizes an onboard Mac mini (M4 Pro) running a Visual Language Model (VLM) and a Diffusion Transformer (DiT) for task orchestration and motion generation, while a dedicated chip runs a Reinforcement Learning (RL) policy network for high-frequency action tracking. Tests were conducted on Unitree G1 robots in six unscanned environments, achieving an average success rate of 91% across 360 single tasks. For complex multi-step tasks, the average success rate was 70%, with a specific 'crouch-obstacle-crossing-push-box' sequence succeeding in 5 out of 10 attempts. The team used 250 H100 GPUs to synthesize training data and noted that current reliance on vision limits performance when objects are occluded or contact is not directly observable.
- Chinese robotics — Hardware and sensingSignalApple Research Discloses HALO Autonomous Control System for Humanoid Robots Validated on Unitree G1
Apple Research disclosed the HALO autonomous control system for humanoid robots, validated on Unitree G1 hardware in six unscanned real-world environments with an average single-task success rate of 91%. The system employs a three-layer architecture: a local visual language model (Qwen3.6-35B-A3B) for task orchestration, a Qwen3.5-2B model for action generation, and a reinforcement learning-based tracker for joint control. Training data was synthesized from 1,000 simulated environments over approximately 300 hours using 250 NVIDIA H100 GPUs.
- Chinese robotics — Research news discoverySignalApple Research Team Discloses Progress on Humanoid Robot 'Brain' System
Apple Research disclosed progress on its humanoid robot autonomous control system, HALO, via an arXiv preprint titled "Perceptive Humanoid Motion from Simulation." The team released code for data generation, model training, and deployment covering action generation, tracking, task orchestration, simulation, and real-robot execution. Tests in six unscanned real-world environments achieved a 91% average success rate across 360 single tasks; combined task tests in one environment averaged 70%. The study acknowledges that simulation data still requires human prompts and motion capture for new physical tasks, and notes limitations in visual-only perception for contact-heavy or occluded-object scenarios.
- arXiv Robotics — research abstractsSignalPerceptive Humanoid Motion from Simulation
This research presents HALO, a humanoid autonomy system that overcomes the data bottleneck for generalist robots by synthesizing vision-language-motion datasets across procedurally generated environments. The system learns diverse physical skills such as sitting, navigating to named objects, and pushing boxes in simulation, then transfers these capabilities zero-shot to real-world settings without ever having seen a real home during training. HALO achieves a 91% success rate across 360 trials in six unmapped real-world environments, demonstrating effective sim-to-real transfer for composite, long-horizon tasks.
Event coverage history
Current attention 19·Peak within the comparable range 19（2026-11-07 11:00 UTC）·Change within the comparable range over 24 hours –
The trend compares participants observed continuously; its scope may be narrower than current attention totals. Hover or click to explore hourly observations, or use the left and right arrow keys.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).