SHOP2: An HTN Planning System (1106.4869v1)

Abstract: The SHOP2 planning system received one of the awards for distinguished performance in the 2002 International Planning Competition. This paper describes the features of SHOP2 which enabled it to excel in the competition, especially those aspects of SHOP2 that deal with temporal and metric planning domains.

• The paper presents a domain-independent HTN planner that uses hierarchical task decomposition and interleaves partially ordered subtasks for enhanced efficiency.
• The paper demonstrates robust performance by solving 899 of 904 competition problems across diverse domains, underscoring its practical impact.
• The paper integrates domain-specific heuristics and external program calls to optimize search and manage complex temporal and numeric planning tasks.

Overview of the SHOP2 HTN Planning System

The paper "SHOP2: An HTN Planning System," authored by Dana Nau and colleagues, describes the features and capabilities of SHOP2, a domain-independent planning system based on Hierarchical Task Network (HTN) planning. SHOP2 received distinguished performance recognition in the 2002 International Planning Competition, demonstrating its capability to handle complex planning tasks efficiently, particularly in temporal and metric planning domains.

Key Features and Capabilities

SHOP2 advances its predecessor, SHOP, by incorporating several enhancements that significantly improve its performance and flexibility. One fundamental improvement is the ability of SHOP2 to handle partially ordered tasks, allowing subtasks from different tasks to be interleaved. This flexibility facilitates a more intuitive specification of domain knowledge and improves planning efficiency. Additionally, SHOP2 supports numerous features from the Planning Domain Definition Language (PDDL), such as quantifiers and conditional effects, improving its expressive power.

Important capabilities of SHOP2 include:

• Axiomatic Inference and Symbolic/Numeric Computations: SHOP2 can make complex inferences and perform computations that involve both symbolic and numeric data.
• External Program Calls: SHOP2 can call external programs to perform tasks that might be computationally intensive or domain-specific.
• Temporal Planning: The system includes a mechanism to translate temporal PDDL operators into SHOP2 operators, ensuring that it can manage multiple timelines and maintain state consistency in temporal planning domains.
• Optimized Search: SHOP2 incorporates a "sort-by" construct to guide the search process toward more promising areas of the search space. This feature is particularly useful in optimization problems where some heuristic can estimate the cost or benefit of pursuing a particular path.

Planning Algorithm

The core of SHOP2's planning mechanism is its planning algorithm, which generates plans by decomposing tasks into subtasks in a hierarchical fashion. This decomposition continues until primitive tasks, which can be directly executed using planning operators, are reached. SHOP2's algorithm ensures that the current world state is known at each step, which simplifies reasoning and reduces uncertainty. This feature distinguishes SHOP2 from classical planners and allows it to represent and solve more complex planning problems.

A key aspect of SHOP2's planning approach is its use of "standard operating procedures" encoded as HTN methods for task decomposition. This makes the planner highly customizable to specific domains, leading to significant efficiency gains and the capability to handle problems that might be intractable for fully automated planners.

Performance and Competition Results

In the 2002 International Planning Competition, SHOP2 demonstrated its robustness by solving a remarkable 899 out of 904 problems across various domains, including Strips, Numeric, HardNumeric, SimpleTime, Time, and Complex domains. This performance highlighted SHOP2's versatility and effectiveness in a wide range of scenarios. The ability to solve more problems than any other competitor in the competition underscores the effectiveness of the HTN planning approach in handling complex, real-world problems.

In terms of speed, SHOP2 generally performed slower than peers like TLPlan and TALPlanner, but showed competitive advantages in certain domains, such as Satellite-HardNumeric. The competition results suggest polynomial relationships in the runtime of these hand-tailorable planners, attributed to their ability to embed domain-specific knowledge, thereby minimizing the need for extensive search and backtracking.

Related Work

The paper contextualizes SHOP2 within the broader landscape of HTN planning. HTN planning, developed over 25 years ago, is well-suited for practical applications such as production-line scheduling, crisis management, spacecraft planning, and more. The formal semantics of HTN planning make it strictly more expressive than classical AI planning, allowing it to represent a broader range of problems.

Moreover, the paper discusses the relationship between SHOP2 and other hand-tailorable planners like TLPlan and TALPlanner. While TLPlan and TALPlanner use temporal logic for search control, SHOP2 relies on HTN methods to focus its search space on viable solutions. This distinction highlights the varied approaches within the HTN planning paradigm and suggests potential avenues for integrating these strategies.

Conclusion

SHOP2 represents a significant advancement in HTN planning, offering enhanced capabilities for dealing with complex, temporal, and metric planning tasks. Its performance in the International Planning Competition underscores its practicality and efficiency. By focusing on domain-specific knowledge and maintaining the current state throughout the planning process, SHOP2 can generate high-quality plans efficiently. Future developments in AI planning might further explore the integration of different types of control knowledge and optimization strategies, building on the foundation laid by systems like SHOP2.