Interaction-aware planning and optimal control strategies for automated parking can be derived by model-based reinforcement learning (MBRL). For MBRL based parking, collision avoidance and performance guarantee benefit from the use of vehicle model integrated in the agent for planning parking trajectory. However, the vehicle model significantly influences the training process and final parking accuracy. To improve the parking performance, a model refined RL based automated parking is proposed using system identificated transfer function model. Vehicle kinematic model, dynamic bicycle model, speed and steering transfer function models approximating the longitudinal and lateral dynamic characters of the chassis control are used for comparison in this study.

To obtain the identificated model, first, the kinematic model is adopted in reinforcement learning, which is implemented with Monte Carlo tree search (MCTS) and neural network (NN), a less precise parking planner is obtained and the order sequence. Second, the real vehicle data are collected to obtain response characteristics of chassis control system using the order sequence. Then, the vehicle model is used to train a reinforcement learning model. The receding horizon search scheme for the motion planning and control is realized via tree search guided by trained NN. The NN learns the experience of tree search. MCTS search stronger moves for parking, resulting in the reinforcement of parking planning and control in the next iteration. In order to speed up the off-line learning process, multithreading parallel computing is employed, which encourages each instance of simulation with different strength of action exploration. A sampling strategy of increasing sampling times on-the-fly is proposed to promote the robustness of the proposed algorithm.

The influence of different vehicle models on the learning process and control is studied by the simulations. It shows that low order transfer function models are preferable for the proposed method. The real vehicle experiment validates the effectiveness of the proposed method.

The development of autonomous vehicles requires extensive testing of software modules. Developing a reliable software platform which allows testing on a real vehicle is yet a challenging task. Therefore, open-source software platforms are becoming more important for researchers in the field of autonomous driving. For example, Baidu provides the open-source autonomous driving platform Apollo which aims to accelerate testing and deployment of autonomous vehicles. However, the complex software structure hinders an easy integration of developed software modules, especially the motion planning module. Moreover, Baidu's Apollo provides only one possibility to test one's own algorithms in simulation, namely to upload the algorithm in Baidu's cloud platform, which is unacceptable for most autonomous driving companies.

In contrast, the open-source CommonRoad benchmark suites contain diverse testing scenarios, e.g., highway, urban, dense traffic, and interaction with bicyclists and pedestrians. In addition, CommonRoad provides a motion planning framework in Python which enables rapid planner prototyping, along with additional tools, e.g., efficient collision checker, map format converter, and interface with the traffic simulator SUMO.

In this work, we introduce a Python API between the planning module of the Baidu Apollo platform and the CommonRoad software framework. The developed interface aims to bridge the gap between rapid prototyping for safe planning algorithms and real-time test drives. The API transfers perception and map information to the planner and then returns the planned trajectory. The users can either replace the Apollo planner with their own planner or integrate their planner as a fail-safe planner if the planned trajectory by Apollo is unsafe. With our interface, developers can first test their planners in diverse scenarios from the CommonRoad benchmark, and directly on a real vehicle afterwards using the Apollo platform. The latter can be performed without adapting their algorithms to Apollo software structures. Moreover, developers can record their test drives in CommonRoad format for offline analyses. We demonstrate our interface in several scenarios with increasing complexity.

Keywords: Autonomous Driving, Motion Planning, Open-Source Software Framework

Worldwide, regions are on the verge of collapse due to an ongoing urbanization and an increasing number of cars. To relieve overcrowded roads, future urban mobility has to become more flexible and more efficient. The idea of transmodular mobility is to change and share different transportation carriers within a higher-level transportation system. This enables and combines the efficient use of different carrier systems like railways, buses, ropeways or aircraft and increases passenger convenience. Practically, this would mean that different transportation carriers share the same kind of passenger cabins. For instance, a passenger cabin can perform as a bus to reach small and narrow streets and for particular tracks, it would transform into an urban ropeway to benefit from its higher passenger throughput. However, the critical part of this transformation is the coupling process. Therefore, reliable coupling interfaces are required which must be precisely aligned to perform a successful coupling procedure. In this regard, reliable sensor systems are required which can provide precise positioning data of the coupling interfaces for their counterparts.

Conventionally, automotive sensors for range detection purposes either consist of radar or sonar-based systems. Automotive radars are known for a comparable large detection range and can cover up to hundreds of meters. However, the resolution is typically in the range of a few centimeters and even more. Sonar-based systems feature less detection range and are mostly used for short-range detection of obstacles in direct proximity to the sensor. Due to their high resolution and low production costs, they are used for a wide range of park assist systems. Nevertheless, what conventional radar and sonar-based sensor systems have in common is that they do not distinguish between particular obstacles without further signal processing. This means, objects around the sensor can be detected but cannot be identified. However, object identification can be achieved by more advanced systems like LIDAR, camera-based or radar-based imaging techniques. This, in turn, comes with an increasing amount of signal processing and production costs.

This paper presents a new concept of a positioning sensor system for transmodular mobility without the need for complex image and signal processing capability. It is based on a sonar system approach, in which the transmitting nodes and receiving nodes are separated. Thereby, the transmitting node can serve as a reference point on the target object (e.g. passenger cabin) and the receiving node can be placed on the locating object (e.g. transportation carrier), respectively. In this regard, both objects are related to each other in a local reference system. The sensor system is extended by an auxiliary radio link for synchronization purposes. It can cover a range of more than 5 meters with a maximum positioning resolution in the millimeter range.