Enhanced LGMD Model with Adaptive Probabilistic Regulation for Compound Interference

Area of Science:

• Robotics
• Computer Vision
• Biomimetic Systems

Background:

• Current LGMD-inspired collision detection models suffer from false alarms and performance degradation due to spatial noise and high-frequency jitter.
• Robust visual perception is crucial for micro-robot navigation and collision avoidance.

Purpose of the Study:

• To propose an enhanced visual perception model for micro-robots that overcomes limitations of current collision detection systems.
• To improve the robustness and reliability of LGMD-inspired visual systems under compound interference.

Main Methods:

• Incorporation of adaptive Gaussian random variables to filter spatial noise.
• Integration of spatial residual feedback (SRF) to suppress image shifts from jitter.
• Validation using synthetic and real-world video sequences.

Main Results:

• The proposed model significantly reduces false responses and perceptual degradation under compound interference.
• Maintained robust performance metrics including success rate (SR), discrimination ratio (DR), and membrane potential stability index (MPSI).
• Ablation studies confirmed the synergistic effect of SRF and adaptive Gaussian variables.

Conclusions:

• The enhanced model provides robust collision perception for micro-robots operating in noisy and unstable environments.
• Adaptive probabilistic configurations outperform fixed ones under compound interference.
• This work increases the practical applicability of LGMD-inspired visual systems in real-world scenarios.