Core Concepts
The TCS-AD stack provides a modular and adaptable architecture that enables the integration and evaluation of novel autonomous driving algorithms across a variety of vehicle platforms and real-world environments.
Abstract
The paper presents the TCS-AD stack, a modular and scalable architecture for automated driving functions. The key aspects are:
Modular design with well-defined interfaces to enable easy integration and testing of new components and algorithms.
Flexibility to be deployed on different vehicle platforms, including modified EasyMile EZ10 shuttles and passenger cars, with varying sensor setups, control systems, and performance characteristics.
Deployment and evaluation in real-world environments, including urban areas with mixed traffic, narrow streets, and interactions with vulnerable road users.
The stack includes all necessary components for autonomous operation, such as localization, perception, prediction, planning, and control.
The stack has been extensively tested, with over 3000 km of autonomous driving in the Karlsruhe region of Germany.
The modular design enables the easy integration of machine learning-based algorithms in various parts of the software stack.
The goal is to develop algorithms that can eliminate the need for safety operators, enabling fully autonomous operation.
Stats
The TCS-AD stack has been deployed and tested in real-world environments, with over 3000 km of autonomous driving in the Karlsruhe region of Germany.
Quotes
"The ability to replace components within the software stack and observe their impact on different vehicles and real-world driving conditions has enhanced our research capabilities."
"By swapping out individual components, we can isolate and identify the effect of each component on the overall driving performance."