Abstract:Accurate ego-motion estimation is the foundation for autonomous vehicle navigation in complex urban environments. In multisensor fusion schemes, in addition to environment-related degradations, sensor measurements sometimes fail, which may have an unpredictable impact on navigation performance. To make the vehicle sensor system resistant to different types of failures, we propose a failure-resistant and lightweight multisensor fusion method, in which pseudorange and carrier phase from multiple global navigation satellite system (GNSS) constellations, sparse visual features from a monocular camera, and records from multiple low-cost micro-electromechanical system (MEMS) inertial measurement units (IMUs) are integrated at the measurement level by a sliding-window tightly coupled filter. Furthermore, considering the possibilities of the high-rate IMUs recovering from failures, we propose a flexible strategy to fully utilize all available IMU measurements without frequent reinitializations. To verify the proposed method, we perform extensive evaluations under different sensor failures (e.g., GNSS short-term blocking and long-term failures, image failures, and complex IMU failures and recoveries). The results show that our method outperforms conventional approaches that include only a single IMU and merely focus on sensor degradation, and resists various types of sensor failures. Despite initializing under a harsh environment and experiencing complex sensor failures, velocity and yaw estimations are significantly improved while a submeter 3-D positioning can be achieved in approximately 80% of the challenging scenario. For complex IMU failure and recovery records, our method can fully utilize all available IMU measurements and seamlessly switch between the base and auxiliary IMUs to recover trajectories that approximate the ground truth.

Learning Robust Vehicle Navigation Policy under Interference and Partial Sensor Failures

Graph Neural Network Based Abnormal Perception Information Reconstruction and Robust Autonomous Navigation

Feature Uncertainty Estimation In Sensor Fusion Applied To Autonomous Vehicle Location

Probabilistic End-to-End Vehicle Navigation in Complex Dynamic Environments with Multimodal Sensor Fusion

Neural Network Based Uncertainty Prediction for Autonomous Vehicle Application.

Robust and Transferable End-to-end Navigation Against Disturbances and External Attacks: an Adversarial Training Approach

Anomaly Detection for Robust Autonomous Navigation.

Highly Robust and Accurate Multi-Sensor Fusion Localization System for Complex and Challenging Scenarios

A Failure-Resistant, Lightweight, and Tightly Coupled GNSS/INS/Vision Vehicle Integration for Complex Urban Environments

Fault-Tolerant Multi-Sensor Fusion Positioning System for Autonomous Vehicles in Unknown Outdoor Environments

A Robust Fusion Methodology for MEMS-Based Land Vehicle Navigation in GNSS-Challenged Environments

Multi-modal policy fusion for end-to-end autonomous driving

Deep Reinforcement Learning Based Mobile Robot Navigation Using Sensor Fusion

Adaptive Sensor Fusion of Camera, GNSS and IMU for Autonomous Driving Navigation

Feature-based Visual Navigation Integrity Monitoring for Urban Autonomous Platforms

Learning Resilient Behaviors for Navigation Under Uncertainty

Double Deep Q-Learning and Faster R-CNN-Based Autonomous Vehicle Navigation and Obstacle Avoidance in Dynamic Environment

Navigation Engine Design for Automated Driving Using INS/GNSS/3D LiDAR-SLAM and Integrity Assessment

Robust Multi-Sensor Fusion SLAM Based on Road Network and Reflectivity Enhancement

Convergence of Communications, Control, and Machine Learning for Secure and Autonomous Vehicle Navigation

Learning End-to-end Multimodal Sensor Policies for Autonomous Navigation