用惯性数据同步雷达与激光雷达，降低定位计算成本。
IMU-Preintegrated Radar Factors for Asynchronous Radar-LiDAR-Inertial SLAM
- 用高频率惯性数据将激光雷达状态推算到雷达采样时刻
- 节点数量减少50%，优化时间最多降低56%
- 适合资源受限设备实时运行的多传感器融合
固定滑动窗口的雷达-激光雷达-惯性紧耦合平滑器通常为每条测量创建一个因子图节点，以补偿雷达与激光雷达之间的时间不同步。当雷达和激光雷达采样率相等时，该策略导致状态创建速率是单个传感器频率的两倍，使每秒状态数翻倍，显著增加优化开销，阻碍在资源受限硬件上的实时性能。本文提出惯性预积分雷达因子，利用高频惯性数据将最新的激光雷达状态传播至雷达测量时间戳。该方法将节点创建率维持在激光雷达测量频率水平。在传感器速率相等的假设下，节点数量减少50%，从而降低计算成本。实验基于单板计算机（4核2.2 GHz A73 + 2 GHz A53，8 GB RAM）进行，结果表明，本方法在保持与传统基线相当的绝对位姿误差的同时，因子图优化总耗时最多降低56%。
原文摘要 · Abstract (English)
Fixed-lag Radar-LiDAR-Inertial smoothers conventionally create one factor graph node per measurement to compensate for the lack of time synchronization between radar and LiDAR. For a radar-LiDAR sensor pair with equal rates, this strategy results in a state creation rate of twice the individual sensor frequencies. This doubling of the number of states per second yields high optimization costs, inhibiting real-time performance on resource-constrained hardware. We introduce IMU-preintegrated radar factors that use high-rate inertial data to propagate the most recent LiDAR state to the radar measurement timestamp. This strategy maintains the node creation rate at the LiDAR measurement frequency. Assuming equal sensor rates, this lowers the number of nodes by 50 % and consequently the computational costs. Experiments on a single board computer (which has 4 cores each of 2.2 GHz A73 and 2 GHz A53 with 8 GB RAM) show that our method preserves the absolute pose error of a conventional baseline while simultaneously lowering the aggregated factor graph optimization time by up to 56 %.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。