Bioinspired Stimulus Selection Under Multisensory Overload in Social Robots Using Reinforcement Learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Cognitive Science

Background:

• Autonomous social robots require real-time environmental interpretation via multimodal perception.
• Sensory overload and false positives can degrade robot performance and interaction coherence.
• Current methods struggle with prioritizing relevant stimuli in complex environments.

Purpose of the Study:

• To develop a Bioinspired Attentional System for real-time stimulus prioritization in social robots.
• To address sensory overload and improve the selection of relevant inputs.
• To enhance the quality and coherence of human-robot interactions.

Main Methods:

• Utilized Reinforcement Learning (RL) to manage stimulus prioritization.
• Incorporated neurocognitive mechanisms: Inhibition of Return and Attentional Fatigue.
• Defined an RL reward function to dynamically adjust stimulus weights based on relevance and temporal/modal factors.

Main Results:

• The system effectively modulates sensory signals and reduces the impact of redundant inputs.
• Demonstrated improved stimulus selection in overstimulating scenarios through three case studies.
• Significantly reduced expression queue length and execution delay compared to a baseline queue system.

Conclusions:

• The Bioinspired Attentional System enhances autonomous social robot performance in complex perceptual environments.
• The system effectively manages multimodal sensory input, leading to more coherent interactions.
• This approach offers a promising solution for real-time stimulus management in advanced robotic systems.