TinyLidarNet: 2D Lidar-Based End-To-End Deep Learning Model for F1TENTH Autonomous Racing

Prior research has demonstrated the effectiveness of end-to-end deep learning for robotic navigation, where the control signals are directly derived from raw sensory data. How- ever, the majority of existing end-to-end navigation solutions are predominantly camera-based. In this paper, we introduce TinyLidarNet, a lightweight 2D LiDAR-based end-to-end deep learning model for autonomous racing. An F1TENTH vehicle using TinyLidarNet won 3rd place in the 12th F1TENTH Au- tonomous Grand Prix competition, demonstrating its compet- itive performance. We systematically analyze its performance on untrained tracks and computing requirements for real-time processing. We find that TinyLidarNet’s 1D Convolutional Neu- ral Network (CNN) based architecture significantly outperforms widely used Multi-Layer Perceptron (MLP) based architecture. In addition, we show that it can be processed in real-time on low-end micro-controller units (MCUs).