A Collaborative Parametric Design Agent

Daniel Kuokka, Brian LivezeyAAAI 1994

Introduction

In modern engineering, "parametric design" is a foundational methodology where systems are defined through variables (parameters) and the mathematical relationships (constraints) that govern them. While traditionally a solitary task supported by localized software, complex projects—such as the design of satellites or aircraft—require the coordination of diverse engineering teams. In these environments, a change made by a "Bus" designer regarding frame length can inadvertently violate a weight constraint held by the "Payload" department.

Historically, integrating these disparate constraints was a manual, error-prone process. This paper introduces ParMan (Parametric Manager), a distributed design tool that transforms parametric design into a collaborative team effort by combining constraint logic programming with an agent-based communication infrastructure.

System Architecture and Agent Communication

ParMan operates as a semi-autonomous agent representing a human designer within a network. The architecture is built upon three primary pillars designed to facilitate real-time negotiation and conflict detection.

At the core of the communication layer is the Matchmaker agent. Rather than requiring agents to know the specific addresses of their counterparts, ParMan utilizes content-based routing. Agents "advertise" the parameters they control and "subscribe" to the parameters they need. The Matchmaker facilitates these connections dynamically, allowing the design team to scale or change without reconfiguring the entire network.

The organizational structure of ParMan agents and the Matchmaker facilitating communication between specialized designers.

Technical Implementation

The "math engine" of ParMan is based on Constraint Logic Programming (CLP(R)). Unlike standard calculators, CLP(R) reasons about the feasibility of a set of constraints, determining if a solution space even exists.

Key technical features include:

  • KQML & KIF: The system uses the Knowledge Query and Manipulation Language (KQML) for inter-agent dialogue and Knowledge Interchange Format (KIF) for representing the constraints themselves.
  • Symbolic Reasoning: While early prototypes were limited to numeric values, the authors extended the system to support symbolic constraints (e.g., material_type = aluminum), which are essential for high-level engineering decisions.
  • Automated Unit Conversion: To prevent the common engineering pitfall of mixed-unit errors, ParMan automatically converts all inputs to SI units before processing.

User Interface and Conflict Visualization

ParMan provides a graphical user interface (GUI) that allows engineers to manage complexity without being overwhelmed by the global state of the project. Users define parameters in a hierarchical tree and view them in a specialized "Parameter Table."

One of the system's most critical innovations is its visual feedback for conflict detection. The interface uses a color-coded system to alert designers to inconsistencies:

  • Red Indicators: Signal a local conflict where the user's own constraints are internally contradictory.
  • Yellow Indicators: Signal a global conflict, indicating that while the local design is valid, it violates a constraint set by a teammate elsewhere in the network.
The ParMan interface showing the Parameter Table and the local constraint editor for the FSAT project.

Evaluation and Case Studies

The system was evaluated across three distinct domains to test its flexibility and robustness:

  1. FSAT (Frugal Satellite): A Lockheed project focused on inexpensive, reusable satellite design. ParMan successfully identified "cliques" of related parameters, isolating inconsistencies between bus and payload subsystems.
  2. Bicycle Design: A standard mechanical engineering benchmark used to test geometric and material constraints.
  3. Meeting Scheduling: A non-engineering test case that demonstrated the constraint engine's ability to handle temporal and resource-based logic.

Insights and Future Work

ParMan demonstrates that distributing the design problem reduces the cognitive load on individual engineers. By only surfacing relevant constraints, the tool allows specialists to focus on their domain while the agent infrastructure maintains global integrity.

However, the authors noted several areas for improvement. The 1994-era UI did not support direct manipulation of constraint graphs, and the performance of the CLP(R) solver could potentially become a bottleneck in extremely large, highly interconnected systems. Future research will focus on more extensive field testing with professional engineers to better understand how these collaborative agents integrate into long-term industrial workflows.

Summary updated

BibTeX
@inproceedings{kuokka1994collaborative,
  author = {Daniel Kuokka and Brian Livezey},
  title = {{A Collaborative Parametric Design Agent}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1994},
}
Figure