Bioinspired Cross-Modal Self-Adaptive Machine Intelligence for Event-Driven and Ultrahigh-Precision Underwater...

Area of Science:

• Robotics and Artificial Intelligence
• Materials Science and Engineering
• Sensor Technology

Background:

• Current embodied intelligent agents struggle with natural environmental interactions due to limitations in unimodal sensing and cross-modal coordination, especially in dynamic, contact-rich scenarios.
• Dexterous manipulation of objects, particularly fragile or slippery ones, remains a significant challenge for robotic systems.
• Underwater operations present unique challenges for robots, where visual perception is often limited, necessitating alternative sensing modalities.

Purpose of the Study:

• To develop a highly sensitive, robust, and event-driven tactile interface for improved robotic perception and manipulation.
• To create a bio-inspired visual-tactile fusion (VTF) architecture for enhanced underwater robotic capabilities.
• To establish a closed-loop cross-modal perception-decision system for advanced autonomous decision-making in intelligent agents.

Main Methods:

• Fabrication of a novel fabric-based tactile interface with a woven structure and cross-fiber electrodes, enabling event-driven, dual-mode pressure/slip feedback.
• Development of a bio-inspired visual-tactile fusion (VTF) architecture integrating tactile data with visual semantics for decision-level optimization.
• Implementation of a closed-loop system linking environmental interaction, cross-modal perception, and autonomous decision-making for humanoid machine intelligence.

Main Results:

• The tactile interface demonstrated breakthroughs in sensitivity (246.3 kPa⁻¹) and pressure detection (>450 kPa), along with waterproof robustness.
• Millisecond-level pressure/slip dual-mode feedback enabled self-adaptive grasping, significantly improving the manipulation of delicate objects.
• The VTF architecture achieved 97.7% accuracy in complex underwater tasks, including manipulation of transparent objects and recognition of similar items.

Conclusions:

• The developed fabric-based tactile interface and VTF architecture significantly advance the capabilities of embodied intelligent agents, particularly in dynamic and underwater environments.
• The event-driven tactile feedback and visual-tactile fusion enhance robotic autonomy, adaptation, and dexterous manipulation, crucial for real-world applications.
• This work represents a significant step towards creating intelligent robotic systems with superior perception-decision capabilities for open-world scenarios.