Apple Research Team Discloses Progress on Robot 'Brain' with 91% Average Success Rate in Single-Task Tests
苹果研究团队披露机器人“大脑”进展 单项任务测试平均成功率为91%
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.
Source: Chinese robotics — Research news discovery · Read original article ↗
Article text · Machine translation into English
On November 7, according to news from The Paper, the Apple research team disclosed research progress on the robot “brain.”
On November 4 local time, the Apple research team submitted the paper “Perceptive Humanoid Motion from Simulation” to the preprint platform arXiv, disclosing the humanoid robot autonomous control system HALO. The research team also released the data generation, model training, and deployment code, covering aspects such as motion generation, motion tracking, task orchestration, as well as simulation and real robot operation.
According to the paper disclosure, the research team conducted 360 single-task tests in 6 real environments that were not pre-scanned or mapped, with an average success rate of 91%. Additionally, in one real environment, 10 tests were conducted for each of 3 combination tasks, with an average success rate of 70%.
Among them, the success rate of the combination “bend down to pass the obstacle—walk towards the target—push the box” is 50%. It should be pointed out that the simulation-generated data did not completely eliminate human participation.
The team stated that adding new physical tasks still requires manual prompt writing and designing action keyframes, and some actions still used human videos or motion capture materials. The research team also pointed out that robots currently mainly rely on visual perception and find it difficult to directly perceive contact situations; when objects leave the camera's field of view or are obstructed, the performance of tasks involving a lot of contact will decrease. Adding tactile perception is one of the directions for subsequent improvements.
▼
Original title: “Apple Research Team Discloses Progress on Robot “Brain” Average Success Rate for Single Task Tests is 91%”
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
task tests
360
Reported trials
combination task tests
10
Reported trials
Reported success rate
91%
Reported success rate
70%
Reported success rate
50%
Dataset details reported by the source
- Embodiments
- humanoid Open source S3
- paper: not_reported
- code: Reported pending Open source S4
- weights: not_reported
- dataset: not_reported
- code: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, 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.
研究团队同时指出，目前机器人主要依赖视觉感知，难以直接感知接触情况；当物体离开相机视野或被遮挡时，涉及较多接触的任务表现会下降。
Open source S10
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 S5
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 S6
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 S3
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 S7
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 S4
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 — Research news discovery · thepaper.cn