A5001063593- Autonomous Vehicle Technology and Safety
- Advanced Neural Network Applications
- Real-time simulation and control systems
- Robotics and Sensor-Based Localization
- Vehicular Ad Hoc Networks (VANETs)
- Robotic Path Planning Algorithms
- Vehicle Dynamics and Control Systems
- Advanced Optical Sensing Technologies
- Gaussian Processes and Bayesian Inference
- Traffic control and management
- Video Surveillance and Tracking Methods
- Domain Adaptation and Few-Shot Learning
- Infrared Target Detection Methodologies
- Adversarial Robustness in Machine Learning
- Software Testing and Debugging Techniques
- Vehicle emissions and performance
- Air Quality Monitoring and Forecasting
- CCD and CMOS Imaging Sensors
- Target Tracking and Data Fusion in Sensor Networks
Technical University of Munich
2021-2023
Technical University of Darmstadt
2019
Abstract For decades, motorsport has been an incubator for innovations in the automotive sector and brought forth systems, like, disk brakes or rearview mirrors. Autonomous racing series such as Roborace, F1Tenth, Indy Challenge (IAC) are envisioned playing a similar role within autonomous vehicle sector, serving proving ground new technology at limits of systems capabilities. This paper outlines software stack approach TUM Motorsport team their participation IAC, which holds two...
Reliable detection and tracking of surrounding objects are indispensable for comprehensive motion prediction planning autonomous vehicles. Due to the limitations individual sensors, fusion multiple sensor modalities is required improve overall capabilities. Additionally, robust essential reducing effect noise improving state estimation accuracy. The reliability vehicle software becomes even more relevant in complex, adversarial high-speed scenarios at handling limits racing. In this paper,...
LiDAR object detection algorithms based on neural networks for autonomous driving require large amounts of data training, validation, and testing. As real-world collection labeling are time-consuming expensive, simulation-based synthetic generation is a viable alternative. However, using simulated the training leads to domain shift testing due differences in scenes, scenarios, distributions. In this work, we quantify sim-to-real by means detectors trained with new scenario-identical dataset....
While current research and development of autonomous driving primarily focuses on developing new features algorithms, the transfer from isolated software components into an entire stack has been covered sparsely. Besides that, due to complexity stacks public road traffic, optimal validation is open problem. Our paper targets these two aspects. We present our vehicle EDGAR its digital twin, a detailed virtual duplication vehicle. vehicle's setup closely related state art, valuable...
Simulation-based testing is seen as a major requirement for the safety validation of highly automated driving. One crucial part such test architectures are models environment perception sensors camera, lidar and radar sensors. Currently, an objective evaluation comparison different modeling approaches automotive still challenge. In this work, real sensor system used object recognition first functionally decomposed. The resulting sequence processing blocks interfaces then mapped onto...
Autonomous vehicles demand high accuracy and robustness of perception algorithms. To develop efficient scalable algorithms, the maximum information should be extracted from available sensor data. In this work, we present our concept for an end-to-end architecture, named DeepSTEP. The deep learning-based architecture processes raw data camera, LiDAR, RaDAR, combines in a fusion network. output network is shared feature space, which used by head networks to fulfill several tasks, such as...
Connected and autonomous vehicles (CAVs) could reduce emissions, increase road safety, enhance ride comfort. Multiple CAVs can form a CAV platoon with close inter-vehicle distance, which further improve energy efficiency, save space, travel time. To date, there have been few detailed studies of self-driving algorithms for platoons in urban areas. In this paper, we therefore propose architecture combining the sensing, planning, control an end-to-end fashion. Our multi-task model switch...
Perception algorithms for autonomous vehicles demand large, labeled datasets. Real-world data acquisition and annotation costs are high, making synthetic from simulation a cost-effective option. However, training on one source domain testing target can cause shift attributed to local structure differences, resulting in decrease the model's performance. We propose novel adaptation approach address this challenge minimize between simulated real-world LiDAR data. Our adapts 3D point clouds...
The objective of this work is to provide a comprehensive understanding the development autonomous vehicle perception systems. So far, most autonomy research has been concentrated on improving systems' algorithmic quality or combining different sensor setups. In our work, we draw conclusions from participating in Indy Autonomous Challenge 2021 and its follow-up event Las Vegas 2022. These were first head-to-head racing competitions that required an entire pipeline perceive environment...
For decades, motorsport has been an incubator for innovations in the automotive sector and brought forth systems like disk brakes or rearview mirrors. Autonomous racing series such as Roborace, F1Tenth, Indy Challenge (IAC) are envisioned playing a similar role within autonomous vehicle sector, serving proving ground new technology at limits of capabilities. This paper outlines software stack approach TUM Motorsport team their participation Challenge, which holds two competitions: A...
Autonomous vehicles demand high accuracy and robustness of perception algorithms. To develop efficient scalable algorithms, the maximum information should be extracted from available sensor data. In this work, we present our concept for an end-to-end architecture, named DeepSTEP. The deep learning-based architecture processes raw data camera, LiDAR, RaDAR, combines in a fusion network. output network is shared feature space, which used by head networks to fulfill several tasks, such as...
Motorsport has always been an enabler for technological advancement, and the same applies to autonomous driving industry. The team TUM Auton-omous Motorsports will participate in Indy Autonomous Challenge Octo-ber 2021 benchmark its self-driving software-stack by racing one out of ten Dallara AV-21 racecars at Indianapolis Motor Speedway. first part this paper explains reasons entering vehicle race from academic perspective: It allows focusing on several edge cases en-countered vehicles,...