L2 Goal Directed
Level 2 embodied AGI adds goal-directed behavior, where agents pursue specified objectives using simple planning or learned policies instead of just reacting to immediate stimuli.
At this level, the robot is no longer limited to repeating the same fixed action. It can work toward a clear goal, such as “pick up the red cup and place it on the table,” and sequence multiple steps to achieve it.
Improvements
L2 systems show noticeably better integration of perception and action. They can adapt somewhat to variations in object position, lighting, or minor environmental changes. Simple planning allows them to break down a task into smaller steps and recover from basic failures, such as retrying a grasp if the first attempt slips.
While still constrained, these agents demonstrate the beginning of purposeful behavior rather than purely reflexive responses. They start to show early signs of compositionality — combining basic skills into short sequences.
Examples
Good examples include modern warehouse robots that can pick items from shelves and place them into bins according to an order, or home assistants that perform short sequenced chores like “fetch the remote and bring it to the couch.” Many current mobile manipulators in research labs and early commercial deployments operate at this level, handling structured but slightly varied tasks with limited flexibility.
Further Learning Resources
- Toward Embodied AGI: A Review of Embodied AI and the Road Ahead (Wang et al., 2025) – The paper that introduced the five-level taxonomy for embodied AGI, including the description of Level 2 capabilities
The Future: Reliable Task Execution
As L2 systems mature, they will become dependable partners in structured environments such as warehouses, factories, and well-organized homes. Robots will reliably complete multi-step tasks with minimal supervision, adapting to small daily variations while maintaining safety and efficiency.
This reliability is a critical stepping stone. Once goal-directed behavior becomes consistent and trustworthy, it paves the way for broader autonomy at higher levels — where agents can handle more open-ended tasks, longer time horizons, and less predictable environments.
Strong L2 performance will reduce the need for constant human oversight in many practical settings and provide the stable foundation needed to safely develop the more advanced cognitive and adaptive capabilities required for Levels 3 through 5. In this sense, L2 represents the transition from basic reactive machines to genuinely useful, goal-oriented embodied agents.
