Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TinyLidarNet: 2D LiDAR-based End-to-End Deep Learning Model for F1TENTH Autonomous Racing

Mohammed Misbah Zarrar, Qitao Weng, Bakhbyergyen Yerjan, Ahmet Soyyigit, Heechul Yun

TL;DR

TinyLidarNet’s 1D Convolutional Neural Network (CNN) based architecture significantly outperforms widely used Multi-Layer Perceptron (MLP) based architecture and it is shown that it can be processed in real-time on low-end micro-controller units (MCUs).

Abstract

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. However, 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 Autonomous Grand Prix competition, demonstrating its competitive performance. We systematically analyze its performance on untrained tracks and computing requirements for real-time processing. We find that TinyLidarNet's 1D Convolutional Neural 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).

TinyLidarNet: 2D LiDAR-based End-to-End Deep Learning Model for F1TENTH Autonomous Racing

TL;DR

TinyLidarNet’s 1D Convolutional Neural Network (CNN) based architecture significantly outperforms widely used Multi-Layer Perceptron (MLP) based architecture and it is shown that it can be processed in real-time on low-end micro-controller units (MCUs).

Abstract

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. However, 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 Autonomous Grand Prix competition, demonstrating its competitive performance. We systematically analyze its performance on untrained tracks and computing requirements for real-time processing. We find that TinyLidarNet's 1D Convolutional Neural 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).

Paper Structure

This paper contains 14 sections, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. F1TENTH Autonomous Racing Competition
  4. End-to-end Deep Learning for Autonomous Driving
  5. F1TENTH Platform and 2D LiDAR
  6. TinyLidarNet
  7. Architecture
  8. Data Collection, Pre-processing and Training
  9. Evaluation
  10. Insights from an F1TENTH Competition
  11. Performance on Simulated Tracks
  12. Inference Latency
  13. Performance on Unseen Real Tracks
  14. Conclusion

Figures (7)

  • Figure 1: 12th F1TENTH Grand Prix Competition Track
  • Figure 2: F1TENTH platform with a Hokuyo UST-10LX 2D LiDAR and a NVIDIA Jetson Xavier NX on-board computer.
  • Figure 3: TinyLidarNet architecture: 9 layers (5 convolutional, 4 fully-connected) with 220,686 parameters.
  • Figure 4: Simulation Tracks from F1TENTH gym and F1TENTH racetrack repositoryokelly2020f1tenthBetz2022_RacingSurvey
  • Figure 5: Speed comparison of different models on GYM Track
  • ...and 2 more figures