Mixed-Integer Motion Planning on German Roads Within the Apollo Driving Stack

Traffic situations with interacting participants pose difficulties for today's autonomous vehicles to interpret situations and eventually achieve their own mission goal. Interactive planning approaches are promising solutions for solving such situations. However, most approaches are only assessed in simulation, as researchers lack the resources to operate an autonomous vehicle. Likewise, open-source stacks for autonomous driving, such as Apollo, provide competitive and resource-efficient state-of-the-art planning algorithms. However, promising planning concepts from research are usually not included within a reasonable time, possibly due to resource restrictions or technical limitations. Without evaluating these novel algorithms in reality, the benefits and shortcomings of proposed approaches cannot be thoroughly assessed. This work aims to contribute methodology and implementation to integrate a novel mixed-integer optimization-based planning algorithm in Apollo's planning component and assess its performance and real-time capability in theory and practice. It discusses the necessary modifications of Apollo for deployment on a different vehicle and presents three real-world driving experiments on a public road alongside a detailed experience report. The driving experiments show a smooth trajectory tracking performance operating robustly under varying perception data quality and the real-time capability of the closed-loop system.