Five AGI Levels
The five levels of embodied AGI provide a useful roadmap for tracking progress toward general physical intelligence. This taxonomy, proposed in the 2025 paper "Toward Embodied AGI: A Review of Embodied AI and the Road Ahead", is inspired by the levels of autonomous driving. It describes a logical progression from narrow, limited assistance to fully autonomous, open-ended, human-like physical capability.
Level 1 offers basic assistance in very limited tasks. Level 5 represents independent performance of diverse, novel real-world tasks with near-human proficiency and adaptability.
Progression Logic
Each level builds on the previous one by increasing several key capabilities at once: the number of sensory modalities the system can handle, cognitive abilities such as planning and reasoning, responsiveness to unexpected events, and the ability to generalize to new situations. Higher levels also demand better hardware, more sophisticated control systems, and richer learning mechanisms.
This stepped progression helps researchers and developers measure progress in a structured way and understand what new capabilities are needed to move from one level to the next.
Current Position
Most embodied systems today operate at early levels — primarily L1 and L2. Level 1 systems can perform simple, repetitive tasks such as basic pick-and-place or vacuuming in controlled environments. Level 2 systems show goal-directed behavior and can handle simple compositional tasks with some flexibility.
Promising advances are appearing in L1–L2 tasks, but true generality, long-horizon planning, and robust performance in completely unstructured environments remain significant challenges.
Further Learning Resources
- Toward Embodied AGI: A Review of Embodied AI and the Road Ahead – The original paper that introduced the five-level taxonomy for embodied AGI
The Future: Climbing Toward L5
Achieving higher levels will require tightly integrated advances across bodies, brains, and learning systems. Success at Level 4 and especially Level 5 could deliver versatile physical agents capable of open-ended tasks with minimal supervision.
These advanced systems would transform daily life, industry, healthcare, and exploration. However, reaching L4–L5 will also raise important safety, ethical, and societal considerations around autonomy, accountability, and human-robot collaboration.
The five-level roadmap gives us a practical way to track progress toward truly general embodied AGI — agents that not only perform tasks but understand, adapt, and act meaningfully in our complex physical and social world.