Apple Research Team Unveils HALO Autonomous Control System for Humanoid Robots
91%成功率！苹果亮出人形机器人大脑HALO 陌生环境直接上手
Apple's research team has published findings on HALO, an autonomous control system for humanoid robots that uses visual language models and motion generation to perform tasks like pushing boxes or sitting in chairs in unfamiliar environments. The system was validated on Unitree G1 hardware in unscanned real-world settings, achieving an average 91% success rate for single tasks and 70% for composite tasks. Ongoing reliance on synthetic data generation and manual prompt engineering for new physical tasks is noted.
Source: Chinese robotics — Hardware and sensing · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Control
- Reported autonomousOpen source S3
Reported numbers
Reported success rate
91%
Reported trials
total_test_trials
360
Reported trials
Reported success rate
70%
Reported trials
Reported success rate
50%
Reported trials
synthetic_training_hours
≈300 hours
View original evidence
为了训练机器人适应不同环境，研究团队建立一套大规模合成资料生成流程，在1,000个程序生成的模拟家庭环境中，制作约300小时的主要训练资料，将机器人看到的画面、接收的指令及对应动作配对。
Open source S20simulation_environment_count
1,000
View original evidence
为了训练机器人适应不同环境，研究团队建立一套大规模合成资料生成流程，在1,000个程序生成的模拟家庭环境中，制作约300小时的主要训练资料，将机器人看到的画面、接收的指令及对应动作配对。
Open source S20gpu_hours_for_synthesis
≈48 hours
View original evidence
在计算资源方面，研究团队使用250张英伟达H100 GPU，耗时约48小时完成主要训练资料的合成；动作生成器采用模仿学习训练，使用64张B200 GPU；动作追踪器则通过强化学习、策略蒸馏等方式训练，使用4张H100 GPU。
Open source S23gpu_count_for_synthesis
250
View original evidence
在计算资源方面，研究团队使用250张英伟达H100 GPU，耗时约48小时完成主要训练资料的合成；动作生成器采用模仿学习训练，使用64张B200 GPU；动作追踪器则通过强化学习、策略蒸馏等方式训练，使用4张H100 GPU。
Open source S23
Dataset details reported by the source
- Embodiments
- G1 Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: Reported pending Open source S4
What remains unknown
Not established in the collected evidence: Environment, Data origin.
Dataset licensing and commercial-use rights are not established here.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
研究重点在于让机器人将模拟环境中学到的动作，应用于未曾事先扫描或建图的真实空间，降低机器人对特定环境的依赖。
Open source S5
该系统通过视觉语言模型（VLM）与动作生成技术，让机器人不仅能理解人类指令，还能依据陌生环境调整动作，自主完成推箱子、坐椅子及穿越障碍物等任务。
Open source S3
为了训练机器人适应不同环境，研究团队建立一套大规模合成资料生成流程，在1,000个程序生成的模拟家庭环境中，制作约300小时的主要训练资料，将机器人看到的画面、接收的指令及对应动作配对。
Open source S20
数据生成流程分为几个阶段：首先依照文字指令产生动作序列，再将动作放入虚拟房间，最后通过模拟工具渲染机器人执行任务时看到的画面。
Open source S21
研究团队近日在预印本平台arXiv发表论文《Perceptive Humanoid Motion from Simulation》，介绍HALO系统的设计与实验成果，并将系统部署在宇树科技（688836-CN）的G1人形机器人上进行验证。
Open source S4
在实际测试方面，研究团队选择6个未经事先扫描或建图的真实环境，进行360次单项任务测试，平均成功率达91%。
Open source S17
研究团队另外在单一真实环境中，针对3种复合任务各进行10次测试，平均成功率为70%。
Open source S18
其中，“弯腰通过障碍物、走向目标、推动箱子”这项连续任务的成功率仅50%，显示机器人处理多步骤动作时仍面临挑战。
Open source S19
在计算资源方面，研究团队使用250张英伟达H100 GPU，耗时约48小时完成主要训练资料的合成；动作生成器采用模仿学习训练，使用64张B200 GPU；动作追踪器则通过强化学习、策略蒸馏等方式训练，使用4张H100 GPU。
Open source S23
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Check the exact dataset version, license, modalities and task coverage against the intended use.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：Chinese robotics — Hardware and sensing · news.mydrivers.com