Advancing Autonomous Vehicle Safety: Integrating Functional Safety and Cutting-Edge Path Planning Algorithms for Enhanced Control
Abdullah bin Hassan
Department of Engineering, Universiti Malaysia Sabah, Jalan UMS, Kota Kinabalu, Sabah, Malaysia
Sukritindra Soni
CVM University , Mechanical Engineering Department
Keywords: Autonomous vehicle (AV), path planning (PP), Functional Safety, Bayesian optimization (BO), Enhanced Spatiotemporal Multi-level LSTM network (ESM-LSTM)
Abstract
Automated vehicles, also referred to as self-driving cars, have attracted considerable attention for their potential advantages in terms of safety, efficiency, and convenience. Nevertheless, their widespread adoption necessitates a thoughtful evaluation of potential hazards and the extent of automation. This paper addresses two pivotal facets related to autonomous driving: functional safety and route planning. In the realm of functional safety, this paper explores the concepts of Functional Safety Concept (FSC) and Safety of the Intended Functionality (SOTIF) within the context of designing autonomous driving systems. The objective is to mitigate risks stemming from scenarios where drivers misuse the technology. This is achieved through the incorporation of fail-safe design principles and a comprehensive consideration of possible system failures to ensure a secure environment for autonomous driving. Path planning stands as a foundational technology for autonomous navigation, and crafting a robust and universally applicable path planning algorithm remains a significant challenge. To tackle this challenge, we introduce an Enhanced Spatiotemporal Multi-Level LSTM (ESM-LSTM) network paired with Bayesian Optimization (BO). This path planning system, rooted in deep learning, leverages Convolution Multi-Level Long-Short Term Memory (Conv-MLSTM) to extract concealed features from sequential image data. The spatiotemporal information is further processed using a 3D Stacked Convolution Neural Network (3D-SCNN) in conjunction with BO. The resultant model furnishes real-time insights, ensuring reliable and precise visual outcomes for autonomous vehicles. The planned trajectory is subject to filtering to assess potential hazards, and a system-safe state architecture is proposed to enhance the control of autonomous vehicles.
