AI100 report highlights embodied AI as a grand challenge
The One Hundred Year Study on Artificial Intelligence (AI100) is a longitudinal research project that assesses how AI develops and how it affects society. The project releases a major report every five years, and the 2026 edition is the third in the series. This year’s report is titled “AI Goes Mainstream: The Scaling Era and the Rise of Generative AI.” The series is designed to track long-term trends rather than short-term hype, offering a stable view of AI’s trajectory for researchers, policymakers, and the public. Each edition builds on the previous one.
The study panel is an international committee of experts. Its members examine AI’s evolution across technical, ethical, economic, and societal domains. The panel responds to 12 standing questions about AI’s advances, opportunities, dangers, and societal effects. It also compares developments since 2021 with the previous panel’s assessments. This comparative structure helps distinguish genuine advances from incremental changes and identifies both continuity and change in the field.
The 2026 AI100 Study Panel report examines a wide range of issues. It looks at the scaling of data and computing resources behind generative AI, as well as advances in scientific discovery. It considers investment in AI infrastructure and the concentration of frontier AI development. It also addresses cybersecurity and military applications, the trustworthiness of digital media, emerging risks, public participation, and international governance. Together, these topics show how AI is becoming embedded in many parts of society and why governance debates are increasingly urgent. The report’s scope reflects the growing scale and influence of AI systems.
One of the standing questions concerns open grand challenges in the field. The experts highlight six challenges as particularly inspiring and scientifically important. The most relevant to the robotics community is the problem of physical AI and embodied intelligence. This challenge sits at the intersection of perception, action, and learning in real-world environments, where uncertainty and variability are unavoidable.
In the five years since the last AI100 report, published in 2021, the field has seen significant progress. Natural language processing (NLP) has been integrated into robotic systems. Vision-language-action models have been used for training and control. Simulation technology has also advanced, allowing models trained in virtual settings to be deployed on physical hardware. These techniques allow robots to follow instructions, adapt to new tasks, and improve through simulated experience, though important obstacles remain. These developments have narrowed the gap between digital AI and embodied systems.
The 2026 AI100 report notes that China is the leader in industrial robot development. China deploys more robots than the rest of the world combined. It also produces about 80% of the world’s humanoid robots. Although this sector currently accounts for just 2% of the market, it is growing rapidly. The report’s observations point to a shifting geography of robotics manufacturing and deployment and suggest that humanoid robotics may be the next area of rapid expansion. The figures underscore China’s central role in robotics supply chains.
Despite the progress of recent years, there remains a huge gap between the demonstration of capabilities and deployment. Many success stories have been in highly constrained environments, such as warehouses. Others involve very specific products, such as floor cleaners. The key challenge remains the sheer complexity of the real world. Constrained settings simplify perception and control, while open environments introduce many unmodeled variables. Bridging this gap requires advances in both hardware and software. These constraints limit generality and make deployment costly.
Several scientific problems must be overcome to close this gap. These include long-horizon planning under uncertainty, achieving dexterous manipulation of diverse objects, and establishing formal performance guarantees. Some practitioners believe that current statistical learning paradigms are sufficient for attacking these open problems. Many others feel that fundamentally new ideas will be required to bring robot deployment into genuine real-world scenarios. The debate reflects a broader question about whether scale alone can solve embodied intelligence. The next five years will show whether new paradigms emerge. How this plays out will be interesting to observe. The full report can be read here.