Hippocampus experience inference for safety critical control of unknown multi-agent linear systems

Area of Science:

• Control Engineering
• Artificial Intelligence
• Computational Neuroscience

Background:

• Risk mitigation for safety-critical control typically relies on simulations, but controllers require adjustments when migrated to real-world systems due to simulation constraints.
• Bridging the gap between simulated and real-world performance is crucial for reliable safety-critical control systems.

Purpose of the Study:

• To design an experience inference algorithm for safety-critical control of unknown multi-agent linear systems.
• To leverage neuro-inspired mechanisms for improved transfer learning and decision-making in control systems.

Main Methods:

• The proposed approach models the hippocampus as a stable linear system using adaptive dynamic programming (ADP) for optimal performance.
• The neocortex and striatum are simultaneously developed using an actor control policy algorithm for real-world system experience inference.
• The algorithm is inspired by the functional relationship between the hippocampus, neocortex, and striatum in the brain for transfer learning and decision-making.

Main Results:

• The developed algorithm facilitates experience inference from simulated to real-world multi-agent linear systems.
• Neuro-inspired modeling enhances the controller's ability to adapt to real-world system dynamics.
• Experimental and simulation studies validate the effectiveness of the proposed approach.

Conclusions:

• The experience inference algorithm offers a novel solution for adapting controllers from simulation to real-world safety-critical applications.
• The brain-inspired architecture provides a robust framework for handling unknown system dynamics and improving decision-making.
• This research contributes to advancing the field of safety-critical control through intelligent system design.