Planning Scope, Horizon, and Granularity

Planning Scope, Horizon, and Granularity are fundamental concepts in the design and operation of AI planning systems and agents. They collectively define the boundaries, temporal extent, and level of detail with which an AI agent formulates plans and executes tasks to achieve its goals. Understanding and properly configuring these aspects is crucial for effective task decomposition, resource management, and decision-making within intelligent systems.

Planning Scope

Planning Scope refers to the breadth and boundaries of the planning process, defining what elements, actions, states, and goals are included or excluded during plan formation. It determines the context within which the AI agent considers the environment, constraints, and objectives.

• Spatial Scope: This defines the physical or logical environment considered during planning. For example, an autonomous delivery robot might limit its planning scope to a single building or a specific neighborhood rather than an entire city.

• Goal Scope: It specifies which goals or subgoals the agent is planning for at a given time. Planning scope can be narrow, focusing on immediate subgoals, or broader, encompassing multiple related goals.

• Resource Scope: This includes the resources (time, energy, computational capacity) the agent accounts for during planning. A broader scope might consider long-term resource allocation, whereas a narrow scope focuses on immediate resource availability.

The planning scope controls problem complexity by delimiting the portion of the environment and objectives the agent models, thereby affecting the feasibility and efficiency of planning.

Planning Horizon

Planning Horizon defines the temporal extent into the future over which the AI agent plans its actions. It determines how far ahead in time the consequences of actions are evaluated and how long the plan extends before re-planning or execution.

• Short Horizon: Plans cover a limited future timeframe, emphasizing immediate or near-term actions. This is beneficial in dynamic environments where conditions change rapidly, requiring frequent plan updates.

• Long Horizon: Plans encompass a longer future period, allowing for strategic, high-level decisions and foresight. This approach is useful when the environment is relatively stable or when long-term goals must be achieved.

• Rolling Horizon: Some systems use a moving or rolling horizon, repeatedly planning over a fixed future window that advances as time progresses. This balances foresight and adaptability.

The choice of planning horizon impacts computational complexity, plan optimality, and responsiveness. Longer horizons may yield more globally optimal plans but require more computation and may be less adaptable to change.

Planning Granularity

Planning Granularity refers to the level of detail and abstraction at which the planning process represents actions, states, and goals. It affects how finely the tasks are decomposed and how detailed the plan steps are.

• Coarse Granularity: Plans are composed of high-level actions or macro-steps, encapsulating sequences of lower-level actions. This abstraction simplifies planning and reduces computational load but may lose detailed control and flexibility.

• Fine Granularity: Plans include detailed, low-level actions and state representations. This allows precise control and adaptability but increases complexity and planning time.

• Hierarchical Planning: Many AI systems employ hierarchical task networks (HTNs) or multi-level planning where coarse-grained high-level plans are recursively decomposed into finer-grained subplans. This hierarchical granularity balances scalability and detail.

Granularity influences the agent's ability to handle complexity, adapt to changes, and utilize knowledge efficiently.

Interrelation of Scope, Horizon, and Granularity

These three dimensions are interdependent and must be balanced carefully:

• A large planning scope combined with a long planning horizon and fine granularity can lead to combinatorial explosion, making planning intractable.

• Restricting the scope or horizon or increasing abstraction (coarser granularity) reduces complexity but may sacrifice plan optimality or adaptability.

• Adaptive systems often dynamically adjust these parameters based on environmental feedback, task urgency, or computational resources.

Practical Implications in AI Agent Design

• Efficiency: Properly limiting planning scope and horizon while choosing appropriate granularity ensures timely planning and response.

• Robustness: Adjusting horizon and granularity can help agents cope with uncertainty and dynamic changes by enabling frequent replanning at appropriate levels of detail.

• Scalability: Hierarchical granularity and scoped planning allow agents to operate in large, complex environments by breaking down problems into manageable parts.

• Resource Management: Setting scope, horizon, and granularity helps balance computational cost with plan quality, crucial for real-time or embedded AI systems.

Example: Autonomous Vehicle Planning

• Scope: The vehicle plans routes within a geographic area relevant to the destination, ignoring distant regions.

• Horizon: It plans a route several minutes ahead but re-evaluates frequently to respond to traffic or obstacles.

• Granularity: High-level route planning is coarse (waypoints), while low-level control (steering, speed adjustments) is fine-grained, managed by separate control modules.

These concepts collectively empower AI agents to make informed, efficient, and context-appropriate decisions by structuring the planning process in dimensionally manageable and computable ways. Designing planning algorithms requires careful tuning of scope, horizon, and granularity according to the task, environment, and system capabilities.