Combining Specialized Reasoners and General Purpose Planners: A Case Study

Subbarao Kambhampati, Mark R. Cutkosky, Marty Tenenbaum, Soo Hong Lee

This 1991 paper from the AAAI proceedings explores a fundamental tension in Artificial Intelligence: the trade-off between the flexibility of general-purpose planners and the "deep" expertise required for specialized domains like manufacturing. By proposing a hybrid architecture that decouples high-level logical planning from low-level geometric and physical constraints, the authors demonstrate a system capable of generating and incrementally revising complex machining plans.

The Problem: Symbolic Logic vs. Physical Reality

In the early 1990s, a significant gap existed between the abstract logic of AI planners and the practical requirements of physical domains. General-purpose planners often lacked the "deep" knowledge required to account for 3D geometry, tool interferences, or fixturing stability. Conversely, domain-specific algorithms were efficient but lacked the flexibility to handle diverse tasks.

Encoding 3D geometry and force vectors into standard symbolic logic often leads to a "computational explosion." Previous researchers usually chose between fast but narrow domain-specific algorithms or general-purpose planners using oversimplified rules that often resulted in physically impossible plans.

Proposed Architecture

The authors introduced a hybrid model implemented within the NEXT-CUT concurrent design environment. The system architecture consists of several key components that bridge the gap between logic and geometry:

  • The Planner: A hierarchical nonlinear planner (based on NONLIN) that handles high-level logic, such as tool selection and operation ordering.
  • The Geometry Specialist: Utilizes solid modeling to detect interferences. It translates geometric conflicts into ordering constraints (e.g., "Feature A must be machined before Feature B") for the planner.
  • The Fixturing Specialist: Determines how to hold the part (using vises or clamps) and groups operations into "setups" based on the part's orientation.
  • Specialized Interfaces: Communication occurs via an Interaction Graph and a Setup Graph, allowing modules to share constraints without needing a shared internal language.
The hybrid architecture combining a central planner with specialized geometry and fixture reasoners.

Incremental Revision with PRIAR

A critical feature of the system is its ability to handle design changes without starting from scratch. Using the PRIAR framework, the planner performs "conservative" modifications. When a specialist identifies a problem or a designer modifies a part, the system preserves as much of the existing plan as possible, updating only the affected dependencies to avoid costly re-computation.

Case Study: The Pillow-Block

The system was evaluated using a pillow-block, a standard mechanical component used to support a rotating shaft. The evaluation focused on the system's ability to generate a plan for this part and then revise it when the designer moved a hole or changed a dimension.

The pillow-block part model and the resulting multi-step machining plan generated by the system.

Key Insights

  • Abstraction is the Bridge: The planner does not need to process raw 3D CAD data; it only needs the constraints derived from that data (e.g., "Step 5 must happen before Step 2").
  • Three Sources of Order: The paper identifies that plan orderings come from three distinct places: geometric necessity, planner logic, and fixturing efficiency. Distinguishing these allows the system to prioritize "hard" physical constraints over negotiable efficiency preferences.
  • Efficiency through Reuse: By maintaining causal dependencies, the system can surgically update plans when designs change, leading to significant computational savings compared to full replanning.

Conclusion and Future Work

The case study successfully demonstrated that a hybrid architecture can bridge the gap between general-purpose planning and specialized domain expertise. While the current system treats specialist feedback as absolute, future work aims to incorporate "soft" constraints and more parallel, asynchronous communication between modules to further enhance flexibility and performance.

Summary updated

Abstract

Many real-world planning problems involve substantial amounts of domain-specific reasoning that is either awkward or inefficient to encode in a general purpose planner. Previous approaches for planning in such domains have either been largely domain specific or have employed shallow models of the domain-specific considerations. In this paper we investigate a hybrid planning model that utilizes a set of specialists to complement both the overall expressiveness and the reasoning power of a traditional hierarchical planner. Such a model retains the flexibility and generality of classical planning framework while allowing deeper and more efficient domain-specific reasoning through specialists. We describe a preliminary implementation of a planning architecture based on this model in a manufacturing planning domain, and use it to explore issues regarding the effect of the specialists on the planning, and the interactions and interfaces between them and the planner.
BibTeX
@inproceedings{kambhampati1991combining,
  author = {Subbarao Kambhampati and Mark R. Cutkosky and Marty Tenenbaum and Soo Hong Lee},
  title = {{Combining Specialized Reasoners and General Purpose Planners: A Case Study}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1991},
}

Loading related content...

Figure