Understand this story
On November 4, 2026, Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing HALO, an autonomous control system for humanoid robots. According to East Money, HALO was deployed on Unitree's G1 and completed 360 single-task tests across 6 real-world environments that had not been pre-scanned or mapped, achieving an average success rate of 91%; three types of composite tasks were each tested 10 times, with an average success rate of 70%. The system has three layers: the top layer runs Qwen3.6-35B-A3B locally on a Mac mini carried by the robot to perform task orchestration; the middle layer uses Qwen3.5-2B as a frozen encoder together with a DiT decoder to plan full-body motions for roughly the next 5 seconds; and the bottom-layer tracker outputs joint control commands at a frequency of 50 times per second. The team noted that new physical tasks still require manually written prompts and designed keyframes, that the robot relies mainly on visual perception, and that performance degrades when objects leave the field of view or are occluded; adding tactile perception is one direction for future work. A subsequent report by X.PIN stated that Apple researchers open-sourced HALO, with a single-task success rate of 91%, but that the success rate drops markedly when actions are chained together—for example, falling to about 50% when sequentially crawling under an obstacle, walking, and pushing a box.
AI-generated from public reports
Latest development · Report
Anthropic and NVIDIA Release Robot-Related Models on the Same DayDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 10
East Money
Anthropic and NVIDIA Release Robot-Related Models on the Same Day
On November 9 local time, Anthropic released Fable 5.2 and Mythos 5.2. Among them, Fable 5.2 achieved an average success rate of 64.9% on the RoboDojo dual-arm simulated robot benchmark test, compared with 27.6% for the previous generation Fable 5.1; during the same period, Claude Opus 5.5 scored 39.4% and GPT-6 Astra 29.8%. On the same day, NVIDIA launched the robot model GR00T 2 built on the Cosmos platform, as well as the robot learning framework Isaac Lab 3.0; GR00T 2 achieved an average success rate of 71% across 50 tasks with only 1 demonstration per task, rising to 95% with 5 demonstrations.
Cailian Press
Anthropic Releases Fable 5.2, NVIDIA Launches GR00T 2, Both Doubling Down on Robot Control on the Same Day
Anthropic released its new-generation flagship models Fable 5.2 and Mythos 5.2. On the RoboDojo dual-arm simulated robot benchmark, Fable 5.2 achieved an average success rate of 64.9%, up from 27.6% for the previous generation Fable 5.1; over the same period, Claude Opus 5.5 reached 39.4% and GPT-6 Astra 29.8%. On the same day, NVIDIA launched GR00T 2, a robot model built on the Cosmos platform, and released the robot learning framework Isaac Lab 3.0; GR00T 2 can complete 40 of 50 tasks without specialized demonstrations, with an average success rate of 71% when one demonstration is provided per task, rising to 95% with five demonstrations.
Yicai
Unitree Technology's stock price once surged to third place on the A-share market, as multiple catalysts ignited the humanoid robot sector
After the market opened on November 10, A-share humanoid robot concept stocks rose. Unitree Technology at one point surged more than 12% intraday, with its share price exceeding 800 yuan per share, becoming the third most expensive stock on the A-share market, with its total market value still exceeding 300 billion yuan. Catalysts included Anthropic's release of Claude Fable 5.2 and Claude Mythos 5.2, among which Fable 5.2 achieved a 64.9% success rate in tests on the RoboDojo dual-arm robot; Apple's research team validated the HALO system on the Unitree G1, with an average success rate of 91% across 360 tests in 6 real-world environments.
Sina Technology
Unitree Technology's stock price once surged to third place on the A-share market, as multiple catalysts ignited the humanoid robot sector
After the market opened on November 10, A-share humanoid robot concept stocks rose. Unitree Technology at one point during intraday trading climbed more than 12%, with its share price exceeding 800 yuan per share, becoming the third-highest-priced stock on the A-share market, with a total market capitalization still exceeding 300 billion yuan. Catalysts included Anthropic's release of Claude Fable 5.2 and Claude Mythos 5.2, among which Fable 5.2 achieved a 64.9% success rate in the RoboDojo dual-arm robot test, higher than GPT-6 Astra; Apple's research team validated the HALO system on Unitree G1, with an average success rate of 91% across 360 single-task trials.
TMTPost · AGI
Edge AI Daily Morning Brief: Nvidia Under Investigation, Anthropic's $895.4 Billion Compute Commitment and…
On the same day Nvidia's board launched an independent investigation, Google unveiled two eighth-generation TPUs, the TPU 8tx and TPU 8i Max, targeting confidential computing and high memory bandwidth. Anthropic plans to list on Nasdaq before Thanksgiving, with its valuation range lowered to $1.2 trillion to $1.3 trillion, and cumulative purchase and lease commitments of approximately $895.4 billion. OpenAI completed an initial funding round of about $11 billion, corresponding to a valuation of $1.25 trillion, with the total for this round reaching up to $20 billion.
sspai
Morning Brief: Volcano Engine releases Seedream 5.5, Apple uses simulation to train humanoid robots
Volcano Engine released the Doubao image creation model Seedream 5.5, which supports image generation, editing, cutout, and native RGBA transparent asset generation via natural language. Editing a single 2K square image costs 0.28 yuan, 60% cheaper than Seedream 5.0 Pro and nearly 40% faster, with the API launching on Volcano Ark this week. Apple's research team published a paper on arXiv introducing HALO, a humanoid robot control system, tested on a modified Unitree G1, achieving an average success rate of 91% across 360 tests in 6 unseen real-world scenarios. Anthropic released Claude Fable 5.2 and Mythos 5.2 on the same day, with the standard API priced at $8/$40 per million input/output tokens, a 20% price cut compared to 5.1.
Nov 9
X.PIN
Apple researchers open-source humanoid robot control system HALO, with a 91% single-task success rate
Apple researchers have open-sourced a humanoid robot control system called HALO, which achieves a 91% success rate on a single real-world task. But when actions are chained together, the success rate drops noticeably—for example, when sequentially crawling under an obstacle, walking, and pushing a box, the success rate falls to about 50%. The original text also mentions that Ming-Chi Kuo, based on this, believes that large-scale mass production will not come until 2028 at the earliest.
East Money
Apple research team releases humanoid robot control system HALO, based on Unitree G1 and Qwen model
Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the autonomous humanoid robot control system HALO. The hardware uses Unitree Robotics' G1 with a built-in Mac mini (M4 Pro chip, 64GB unified memory) mounted on its back. HALO has three layers of control: the task orchestrator runs the Qwen large model Qwen3.6-35B-A3B by default, updating approximately every 2 seconds; the motion generator reads first-person RGB images at about 3Hz, generating full-body motion plans for the next 5 seconds; the motion tracker sends commands to joint actuators 50 times per second. The team conducted 360 single-task tests in 6 real environments that were not pre-scanned, achieving an average success rate of 91%, and an average success rate of 70% across 3 combinations of long-sequence composite tasks.
East MoneyRelated discussion
East Money joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
East Money
Apple discloses the HALO system for humanoid robots, with an average task success rate of 91% in real environments
Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" on arXiv, disclosing the humanoid robot autonomous control system HALO, which completed 360 single-task tests in 6 real environments that had not been pre-scanned or mapped, with an average success rate of 91%, and 3 combined tasks each tested 10 times with an average success rate of 70%. HALO is deployed on Unitree's G1 and is divided into three layers: the top layer uses Qwen3.6-35B-A3B running locally on a Mac mini carried by the robot for task orchestration; the middle layer uses Qwen3.5-2B as a frozen encoder together with a DiT decoder to plan full-body motions for about the next 5 seconds; and the bottom-layer tracker outputs joint control commands at a frequency of 50 times per second. The team pointed out that adding new physical tasks still requires manually writing prompts and designing keyframes, the robot mainly relies on visual perception, and performance declines when objects leave the field of view or are occluded; adding tactile perception is one of the future directions.
Electronic Technology Application
Apple publishes research results on HALO, an autonomous control system for humanoid robots
Apple's research team published the paper "Perceptive Humanoid Motion from Simulation" on arXiv, unveiling HALO, an autonomous control system for humanoid robots, and validated it on the Unitree G1 robot. HALO splits the control pipeline into three layers: task orchestration, motion generation, and motion tracking, with the top layer running Qwen3.6-35B-A3B locally on a Mac mini mounted on the robot. The team conducted 360 single-task tests across 6 previously unscanned real-world environments, achieving an average success rate of 91%, but the average success rate for 3 composite tasks was only 70%.
Sina Tech
Apple unveils humanoid robot control system HALO, achieving a 91% task success rate in unfamiliar environments
Apple's research team published the paper "Perceptive Humanoid Motion from Simulation" on arXiv, unveiling the autonomous humanoid robot control system HALO and validating it on the Unitree G1. The system has three layers: task orchestration, motion generation, and motion tracking. The top layer runs Qwen3.6-35B-A3B locally on a Mac mini mounted on the robot. In 360 single-task tests across 6 real-world environments that had not been pre-scanned, the average success rate was 91%; the average success rate for 3 composite tasks was 70%, and among them, the continuous task of bending down to pass an obstacle, walking toward a target, and pushing a box achieved only 50%.
Nov 8
East Money
Apple research team discloses humanoid robot autonomous control system HALO, validated on Unitree G1
The Apple research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the humanoid robot autonomous control system HALO and validating it on the Unitree Technology G1 humanoid robot, enabling motions learned in simulation to be executed in unfamiliar real-world rooms. In the first half of 2026, financing in China's embodied intelligence sector reached approximately 43.8 billion yuan, of which embodied brain companies absorbed 22.253 billion yuan, accounting for 50.8%. AgiBot GO-3Preview raised the phone packaging success rate from 50.8% to 98.3%, and XPeng IRON completed the second-generation VLA on-device deployment.
Nov 7
Yicai
Apple research team discloses humanoid robot control system HALO, validated on Unitree G1 with an average success rate of 91%
The Apple research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the autonomous humanoid robot control system HALO and deploying it on the Unitree G1, enabling motions learned in simulation to be executed in unfamiliar real rooms. The team completed 360 single-task tests across 6 real environments that had not been pre-scanned or mapped, achieving an average success rate of 91%; 3 composite tasks were each tested 10 times, with an average success rate of 70%, among which "duck under an obstacle—walk to the target—push the box" was 50%. The research team publicly released the data generation, model training, and deployment code, and noted that adding new physical tasks still requires manually writing prompts and designing motion keyframes; the robot currently relies mainly on visual perception, and incorporating tactile perception is one direction for future improvement.
East Money
Apple research team releases HALO humanoid robot control system, validated on Unitree G1 with an average success rate of 91%
Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing HALO, an autonomous control system for humanoid robots, and deployed it on a Unitree G1 for validation. The system has three layers: a task orchestration module based on Qwen3.6-35B-A3B runs locally on a Mac mini carried by the robot; Qwen3.5-2B serves as a frozen encoder and, together with a DiT decoder, plans full-body motions of about 5 seconds; and a motion tracker trained with reinforcement learning outputs joint commands 50 times per second. The team completed 360 single-task tests in 6 real environments that were not pre-scanned, achieving an average success rate of 91%, and an average success rate of 70% on 3 combined tasks.
Sina Tech
Apple research team discloses humanoid robot system HALO, verified on Unitree G1 with an average success rate of 91%
Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the autonomous humanoid robot control system HALO and verifying it through deployment on a Unitree G1. HALO has three layers: a task orchestration module based on Qwen3.6-35B-A3B runs locally on a Mac mini carried by the robot, Qwen3.5-2B serves as a frozen encoder to plan full-body motions of about 5 seconds, and a reinforcement-learning-trained motion tracker outputs joint commands 50 times per second. The team completed 360 single-task tests in 6 real environments that were not pre-scanned, achieving an average success rate of 91%, and an average success rate of 70% on 3 combined tasks.
The Paper
Apple research team discloses humanoid robot system HALO, with an average success rate of 91% on the Unitree G1
Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the humanoid robot autonomous control system HALO and validating it on the Unitree G1. HALO has three layers: the top layer uses Qwen3.6-35B-A3B to perform task orchestration locally on a Mac mini carried by the robot, the middle layer uses Qwen3.5-2B together with a DiT decoder to plan full-body motions of about 5 seconds, and the bottom-layer motion tracker outputs joint commands 50 times per second. The team completed 360 single-task tests in 6 real environments that had not been pre-scanned, achieving an average success rate of 91%, and an average success rate of 70% on 3 combined tasks.
Heat trend for this story
- Current
- 6
- Period peak
- 33 Nov 9 12:00
- 24-hour change
- -70%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.