基于语义通信的协同感知系统

Cooperative Perception Based on Semantic Communication

相较于单车感知,协同感知具有更广的感知范围,在自动驾驶进行3D目标检测中发挥着越来越重要的作用。通过V2V通信技术,网联自动驾驶车辆可以共享它们的传感信息(LiDAR点云)以实现协同感知。采用重要性图来提取重要语义信息,并提出了一种具有中间融合功能的创新的协同感知语义通信方案。同时,所提出的架构可以扩展到具有挑战性的时变多径衰落信道。为了减轻时变多径衰落引起的失真,采用了结合信道估计和信道均衡的显式正交频分复用模块。仿真结果表明,所提出的模型在各种信道模型上都优于传统的信源信道分离编码。此外,鲁棒性研究表明,对于协同感知,只有部分语义信息至关重要。尽管所提模型仅在一个特定信道上进行了训练,但它能学习到对各种信道模型都具有鲁棒性的语义信息的编码表示,展示了其通用性和鲁棒性。

Compared to single-vehicle perception,cooperative perception provides a wider sensing range,playing a crucial role in 3D object detection for autonomous driving.Utilizing V2V communication technology,connected autonomous vehicles can share their sensor information(LiDAR point clouds)to achieve cooperative perception.We propose an innovative cooperative perception semantic communication scheme that uses importance maps to extract key semantic information and includes intermediate fusion functionality.Additionally,this architecture can be adapted to handle challenging time-varying multipath fading channels.To mitigate distortions caused by such channels,an explicit orthogonal frequency division multiplexing module that combines channel estimation and channel equalization is employed.Simulation results show that the proposed model outperforms traditional source-channel separate coding across various channel models.Robustness studies indicate that only a portion of the semantic information is crucial for effective cooperative perception.Despite being trained on a specific channel,the model learns a robust encoding of semantic information that remains effective across different channel models,demonstrating its generality and robustness.