LASSA Architecture Enables Autonomous Fault-Tolerant Control for Unmanned Underwater Vehicles
Researchers have introduced a novel intelligent control method for Unmanned Underwater Vehicles (UUVs) based on the LASSA architecture, designed to address the challenges of operating in communication-constrained environments. Traditional fault-tolerant systems often rely on rigid, hard-coded rules that struggle with unforeseen errors, while Large Language Models (LLMs) face issues with hallucinations. The proposed LASSA framework integrates an LLM for autonomous reasoning and task replanning with a solver that verifies physical boundary constraints before commands reach actuators. This structure suppresses physically infeasible outputs and ensures verifiable decision-making through a fast-slow dual closed-loop system. Lake experiments demonstrated the system's efficacy under lower-rudder-fault conditions, where it successfully detected trajectory abnormalities, adjusted turning radius from 4m to 12m, and reduced speed to complete the mission without false alarms. This approach balances high-level decision intelligence with real-time control timeliness, marking a significant advancement in autonomous underwater robotics and AI-driven control systems.
Wire timeline
LASSA Architecture Enables Autonomous Fault-Tolerant Control for Unmanned Underwater Vehicles
Researchers have introduced a novel intelligent control method for Unmanned Underwater Vehicles (UUVs) based on the LASSA architecture, designed to address the challenges of operating in communication-constrained environments. Traditional fault-tolerant systems often rely on rigid, hard-coded rules that struggle with unforeseen errors, while Large Language Models (LLMs) face issues with hallucinations. The proposed LASSA framework integrates an LLM for autonomous reasoning and task replanning with a solver that verifies physical boundary constraints before commands reach actuators. This structure suppresses physically infeasible outputs and ensures verifiable decision-making through a fast-slow dual closed-loop system. Lake experiments demonstrated the system's efficacy under lower-rudder-fault conditions, where it successfully detected trajectory abnormalities, adjusted turning radius from 4m to 12m, and reduced speed to complete the mission without false alarms. This approach balances high-level decision intelligence with real-time control timeliness, marking a significant advancement in autonomous underwater robotics and AI-driven control systems.
cs.AI updates on arXiv.org