HALO Research Paper Reports 91% Success in Tests on Unitree Robot
Key Facts
A research paper dated November 4 introduced HALO, a system that uses simulation to develop humanoid robot motion, and its abstract reported a 91% success rate across 360 trials in six previously unmapped real-world environments. The paper says the system was tested on Unitree’s G1 robot, bringing the Chinese company’s hardware into a research experiment linked to the work of Apple’s artificial intelligence and machine learning team. The results show a transition from training in simulated environments to running the system on a real robot, according to the paper and the research project repository. The experiment does not establish that Apple has developed a commercial robotics product, nor does it establish that the test caused Unitree’s share price to move during the session covered by the news. The study’s significance centers on a simulation-trained system’s ability to perform locomotion tasks in unfamiliar real-world environments.
HALO generates vision, language and motion data across 1,000 procedurally generated virtual homes, then trains the robot on a set of physical skills. Tasks include sitting on chairs, navigating to specified objects, moving sideways through half-open doors, crawling and pushing boxes. The paper presents this as a simulation-training process before performance is tested in real-world environments, while Apple’s repository explains how to run the system on a real G1 robot in a closed control loop. The reported success rate covers 360 trials across six previously unmapped environments, according to the paper’s abstract. These figures remain results from the specific research experiment and do not, by themselves, measure commercial adoption or performance at scale across different operating sites.
The study comes as Unitree’s financial indicators diverge between sales growth and profitability pressure in the third quarter. Quarterly revenue rose 70% year on year to 665 million yuan, while net profit attributable to shareholders fell 17.4% to 112.7 million yuan, according to a Reuters report. Net profit came in below analysts’ expectations of 224 million yuan, the same report said. In the first nine months of the year, revenue increased 55.7% to 1.82 billion yuan, and net profit rose 270.5% to 386.7 million yuan. The company attributed the profit surge during that period to an improved gross profit margin and a large one-off share-based payment expense in the prior-year comparable period.
Unitree’s research and development spending rose sharply in the third quarter as the company invested in embodied AI models, computing and its research and development team. Spending reached 127.0 million yuan, up 249.6% from the same period, and accounted for 19.1% of revenue, according to Reuters. The figures place robotics development among the company’s growing investment outlays, while quarterly net profit came in below analysts’ expectations. In share trading, an earlier Reuters report said the stock had risen more than 65% in the two weeks before its October 28 report, after falling about 55% from its August debut. The company issued an announcement on October 21 about unusual trading and warned of elevated valuation risks, while the report linked the earlier rally to an upgrade from JPMorgan and a demonstration of the H2-X robot.
EL7 data show Apple shares closed at $342.37 on November 6, compared with a close of $344.17 on November 5. These prices relate to sessions before the research paper was submitted on November 4 and do not show the market’s response to its results or a link between Apple’s share movement and the test. The available information also does not establish that Unitree shares rose during the session covered by the news or that the research caused a move. The confirmed connection in the study is the use of the Chinese company’s G1 robot to test the HALO research system. The published results therefore focus on transferring skills learned by the system in simulation to real-world tests, rather than announcing a new Apple robotics product.
The broader commercial test, according to a Morningstar analyst quoted by the Associated Press, is whether Chinese and U.S. robotics companies can deliver reliable performance and attractive returns on investment as they scale into industrial and commercial uses. The agency noted that many humanoid robots are still used mainly for demonstrations, performances and research rather than real-world applications. Unitree said proceeds from its public offering would go toward robotics development and expanding its manufacturing base, according to the agency. These details place HALO’s research findings in the context of the question of moving from experimental success to deploying robots in scalable applications. Neither the paper nor the quoted statements specify a timeline for that transition, while performance and returns in industrial and commercial deployment remain the challenge identified by the analyst.