Qualia as control-theoretic constructs for autonomous agents: event phase space as an action-oriented semantic safety...

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Robotics

Background:

• Autonomous systems struggle with open-world complexity due to combinatorial search limitations.
• Biological agents use holistic situational assessments (qualia) for real-time navigation.
• Existing AI architectures lack the capacity for qualia-like processing.

Purpose of the Study:

• To bridge the architectural gap between biological and artificial agents in complex environments.
• To propose a computational framework for qualia-like assessments in AI.
• To validate the proposed framework in a realistic autonomous driving simulation.

Main Methods:

• Defined qualia operationally through assessment structures with geometric coherence, multimodality, and predictive content.
• Introduced the event phase space, a low-dimensional differentiable manifold with learned dynamics.
• Replaced combinatorial search with gradient navigation for O(n) complexity.
• Validated the model in CARLA Parked Vehicle Occlusion scenarios.

Main Results:

• An alarm function derived from the learned vector field achieved an Area Under the Curve (AUC) of 0.812.
• Reduced collision rates by 29.6% in stress scenarios.
• Outperformed Time-to-Collision (TTC)-based pipelines by 60×.
• Operated at 1.9 ms/decision on CPU with only 6,550 parameters.

Conclusions:

• The proposed qualia-inspired AI architecture offers a computationally efficient and effective solution for autonomous systems.
• The event phase space provides a viable alternative to traditional World Models for complex scene understanding.
• This approach significantly enhances safety and performance in real-world autonomous driving applications.