Architectural mechanisms for dynamic changes of behavior selection strategies in behavior-based systems

Area of Science:

• Artificial Intelligence
• Robotics
• Cognitive Architectures

Background:

• Behavior selection is typically a fixed component in behavior-based AI architectures.
• Adapting behavior selection strategies can enhance agent performance in dynamic environments.

Purpose of the Study:

• To demonstrate the benefits of dynamically changing behavior selection mechanisms.
• To categorize existing mechanisms and identify circumstances for dynamic switching.
• To propose a generic architecture for dynamic behavior selection.

Main Methods:

• Categorized existing behavior selection mechanisms along three dimensions.
• Utilized the Activation, Priority, Observer, and Component (APOC) agent architecture framework.
• Developed a generic architecture for integrating and dynamically switching behavior selection mechanisms.
• Conducted extensive simulations and robotic experiments to verify the proposed mechanisms.

Main Results:

• Identified seven circumstances where dynamic behavior selection is beneficial.
• Demonstrated that the proposed generic architecture can integrate various behavior selection mechanisms.
• Verified the utility of dynamic behavior selection through quantitative and qualitative analysis of simulation and robot experiments.

Conclusions:

• Dynamic behavior selection offers significant performance improvements in specific scenarios.
• The proposed generic architecture provides a unified approach for implementing dynamic behavior selection.
• The findings are validated through extensive empirical studies on both simulated and robotic agents.