Skip to content
Chinese robotics — Research news discovery·· 3 days agoSignalEditorial score75

Apple Research Discloses HALO Autonomous Control System for Humanoid Robots Validated on Unitree G1

平均成功率91%!苹果研究团队披露机器人“大脑”进展:调用千问模型在宇树G1上验证

Summary

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.

Source: Chinese robotics — Research news discovery · 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 S5

Reported numbers

  • Reported success rate

    91%

    Reported trials

    View original evidence
    平均成功率91%!
    Open source S3
  • Reported success rate

    70%

    Reported trials

    View original evidence
    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%。
    Open source S17
  • Reported success rate

    50%

    Reported trials

    View original evidence
    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%。
    Open source S18
  • training_hours

    ≈300 hours

    View original evidence
    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。
    Open source S20
  • task_count

    360

    Reported trials

    View original evidence
    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%。
    Open source S16
  • environment_count

    6

    View original evidence
    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%。
    Open source S16
  • simulation_environment_count

    1,000

    View original evidence
    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。
    Open source S20
  • task_frequency

    50

    View original evidence
    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。
    Open source S15
  • paper: not_reported
  • code: Reported pending Open source S7
  • weights: not_reported
  • dataset: 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.

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 S5

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。

Open source S20

平均成功率91%!

Open source S3

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%。

Open source S17

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%。

Open source S18

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%。

Open source S16

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。

Open source S15

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。

Open source S7

Source:Chinese robotics — Research news discovery · news.qq.com

Timezone · UTC

Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).