Humanoid Robot Intelligence May Advance by 2027

A Chinese embodied AI firm called Spirit AI predicts that the cognitive capabilities of humanoid robots will see a major breakthrough by mid-2027. While the hardware for these machines has already advanced to the point where they can sprint and perform backflips, the software that directs their intelligence remains the primary bottleneck. The company argues that the next two years will be critical for industrial applications, but bringing these machines into private homes will take significantly longer.

According to Gao Yang, co-founder and chief scientist at Spirit AI, the industry is currently facing a severe shortage of suitable training data. This constraint means that while robots are becoming more capable in structured environments like factories, they still lack the adaptability required for the unpredictable nature of domestic life. The firm estimates that household deployment is at least eight years away.

Hardware advances outpace software growth

Robot manufacturers have focused heavily on physical capabilities, resulting in machines that can execute complex physical movements with precision. However, Gao Yang notes that the

The company’s robots have achieved a 90% success rate on simple tasks in controlled living-room settings. Despite this progress, fine-motor skills such as unscrewing bottle caps or handling unfamiliar objects remain difficult. This gap between physical agility and cognitive flexibility suggests that the next phase of development must prioritize software improvements over further mechanical enhancements.

Data collection creates practical bottlenecks

Spirit AI relies on real-world data rather than virtual simulations to train its models. This approach is chosen because simulators struggle with flexible and deformable objects, such as electric cables or soft fabrics, which are common in real-life scenarios. By using actual human movements, the company aims to capture the nuances that virtual environments often miss.

To gather this data, Spirit AI employs around 1,000 contractors across China who use wearable sensors to perform repetitive physical tasks. In their Beijing training center, workers repeatedly open refrigerators, unlock safes, and cut vegetables to provide high-quality movement examples. This method allows the robots to learn from a diverse range of actions, although it is labor-intensive and time-consuming compared to digital training methods.

Industrial use precedes home adoption

The firm believes that the next two years will see robots deployed in commercial service settings for simpler tasks. Dozens of Spirit AI’s wheeled humanoids are already operating on production lines at major companies like CATL and JD.com. These industrial environments are structured and predictable, making them ideal for the current level of robot intelligence.

Transitioning to home environments is far more complex due to the variety of objects and unpredictable user interactions. Gao Yang emphasizes that while the company has raised significant capital and grown rapidly, the path to a fully autonomous household assistant is still steep. The reliance on physical data collection means that scaling this technology requires substantial investment in both human labor and infrastructure, highlighting the trade-off between speed of deployment and the quality of learned behaviors.