Adaptive fuzzy systems for backing up a truck-and-trailer

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Control Systems Engineering

Background:

• Automated vehicle maneuvering, particularly for trucks and trailers, presents significant control challenges.
• Fuzzy control and neural network control are advanced AI techniques applicable to complex control tasks.

Purpose of the Study:

• To compare the performance and robustness of fuzzy control systems and neural-network control systems for truck backing maneuvers.
• To investigate adaptive rule generation methods for fuzzy systems using unsupervised learning.

Main Methods:

• Supervised backpropagation learning trained neural network systems for truck backing.
• Fuzzy associative memory (FAM) rules were evaluated under rule removal and sabotage conditions.
• Unsupervised differential competitive learning (DCL) and product-space clustering were used for adaptive FAM rule generation.

Main Results:

• Neural systems performed well but demanded extensive computational resources for training.
• Fuzzy systems maintained performance until over 50% of FAM rules were removed and tolerated sabotage rules.
• DCL rapidly recovered FAM rules, and product-space clustering translated neural system behavior into FAM rules.

Conclusions:

• Both fuzzy and neural control systems offer viable solutions for automated truck backing.
• Fuzzy control systems demonstrate notable robustness to rule degradation and adversarial conditions.
• Adaptive methods like DCL and clustering can effectively generate or approximate fuzzy rules from data.