Apple Discloses HALO System for Humanoid Robots
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%
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.
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 S2
Reported numbers
Reported success rate
91%
Unique elapsed hours
total_tasks
360
Reported trials
Reported success rate
70%
Reported trials
View original evidence
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%。
Open source S8Reported success rate
50%
Reported trials
View original evidence
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%。
Open source S8total_tasks
10
Reported trials
View original evidence
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%。
Open source S8success_count
5
Reported trials
View original evidence
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%。
Open source S8gpu_hours
≈48 hours
- paper: Reported pending Open source S2
- code: not_reported
- dataset: not_reported
- weights: not_reported
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Data origin.
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.
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%
Open source S1
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。
Open source S2
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%。
Open source S8
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。
Open source S9
Source：Chinese robotics — Hardware and sensing · finance.sina.com.cn