Sleep of Things (SoT): A Neuro-Inspired Digital Twin Framework for Lifelong Memory Consolidation in Autonomous Vehicles

Engineering notes

Recent advances in artificial intelligence and Digital Twin (DT) technologies have significantly enhanced the perception, planning, and control capabilities of autonomous driving systems.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Autonomous vehicles (AVs) are expected to operate safely, intelligently, and reliably in dynamic and unpredictable environments. Recent advances in artificial intelligence and Digital Twin (DT) technologies have significantly enhanced the perception, planning, and control capabilities of autonomous driving systems. However, existing DT frameworks primarily focus on real-time monitoring, simulation, and decision support, offering limited capabilities for systematically consolidating driving experiences and continuously refining knowledge during vehicle idle periods. Consequently, the long-term learning and adaptability of autonomous driving systems remain constrained. This paper presents Sleep of Things (SoT), a novel conceptual framework inspired by the biological process of memory consolidation during sleep. The proposed framework extends the conventional role of DTs beyond real-time synchronization by enabling AVs to review, replay, and consolidate historical driving experiences within a virtual environment while the physical vehicle is inactive. Through offline experience replay and knowledge refinement, the DT continuously improves decision-making policies and transfers the updated knowledge to the physical vehicle. The proposed framework comprises four interconnected components: The Physical Autonomous Vehicle (PAV), the Digital Twin Environment (DTE), the Experience Replay Engine (ERE), and the Memory Consolidation Module (MCM). Together, these components establish a closed-loop lifelong learning paradigm that enables continuous knowledge evolution without interrupting real-world vehicle operation. By integrating neuroscience-inspired memory consolidation into DT technology, SoT extends the functionality of conventional DT architectures and provides a conceptual foundation for developing adaptive, self-improving AVs with lifelong learning capabilities.

Links and sources

Comments