Abstract
This paper introduces a formal framework for Qualitative Mechanics (QM), enabling AI systems to reason about rigid body motion and constraints using symbolic logic rather than complex differential equations. By representing directions and shapes through qualitative signs (+, 0, -), the proposed system, ALEX, predicts the behavior of complex mechanical systems, such as clock escapements, without relying on high-fidelity numerical models.
Introduction
While classical mechanics is a mature field for engineering, it has historically remained a "black box" for Artificial Intelligence. For a robot to perform everyday tasks like turning a doorknob, it requires a "common sense" understanding of physical constraints rather than a full physics engine. This research addresses the challenge of formally describing how objects block or push one another using a symbolic spatial representation that accounts for instantaneous motions in multi-body systems.
Methods
The logic-based theory of rigid body interactions is built upon three pillars:
- Spatial Representation: Directions are represented by an ordered set of signs (+, 0, -).
- Qualitative Vector Arithmetic: Rules for adding and multiplying these signs determine if a motion is into or away from a surface.
- Laws of Blocking and Pushing: These laws define how constraints prevent or transfer motion based on surface normals.
The core insight is that the surface normal is the fundamental unit of mechanical constraint. As shown in the figure above, knowing the normal allows the system to identify the half-plane of motion that is "into" an object and therefore blocked. This framework allows for the representation of partial constraints, such as a block that can slide horizontally but is blocked vertically.
Results: The Recoil Escapement
The theory was implemented in a program called ALEX and evaluated using the recoil escapement, a sophisticated clock mechanism. ALEX successfully determined that when the pallet arm contacts the scape wheel, the wheel's momentum causes it to move backward (recoil) rather than stopping the pallet.
This result demonstrates that qualitative logic can capture critical behavioral transitions in machines—moments that are often difficult to model numerically without precise tolerances. The system correctly identifies that the pallet’s momentum "pushes" the wheel backward during contact.
Limitations and Future Work
Current limitations include the assumption of fixed centers of rotation and a focus on instantaneous rather than continuous motion. Future work aims to integrate this framework with "place vocabularies" to describe how constraints evolve over time and to incorporate more detailed dynamical information to resolve qualitative ambiguities that arise from sign-based arithmetic.
Summary updated