Due to energy/latency/environmental constraints, Endge AI can benefit from specialized sensors, s.a. LIDAR and optimized learning algorithms. SNN have been shown to have lower energy requirements and to perform well on LIDAR data classification. However, previous work requires the entire frame to be scanned prior to classification, and the usage of large networks. We show that a discrete-time Recurrent Spiking Neural Network (RSNN) can efficiently classify LIDAR data, in real-time, throughout the scanning process. The KITTI 3D Object Detection Benchmark is processed into a labeled object classification dataset, on which we simulate the scanning process and test various optimization techniques and encoding methods.