Simon Ulbrich

Probabilistic online POMDP decision making for lane changes in fully automated driving

The Stadtpilot project aims at fully automated driving on Braunschweigs inner city ring road. The TU Braunschweigs research vehicle “Leonie” is one of the first vehicles having the ability of fully automated driving in real urban traffic scenarios. This paper shows our decision making approach for performing lane changes while driving fully automated in urban environments. We apply an online Partially Observable Markov Decision Process (POMDP) to accommodate inevitable sensor noise to be faced in urban traffic scenarios. In this paper we propose a two step algorithm to keep the complexity of the POMDP low enough for real-time decision making while driving. The presented approach has been integrated in our vehicle and was evaluated in real urban traffic.

Multi-Target Tracking using a 3D-Lidar sensor for autonomous vehicles

Environmental perception is a prerequisite for autonomous driving and also a challenging task particularly in cluttered dynamic environments such as complex urban situations. In this paper, we present a robust algorithm for Multi-Target Tracking (MTT) using a Velodyne 3D HDL-64 Lidar sensor. The main contribution of this paper is a practical framework for selecting and representing useful information from the sensor raw data. Since the sensor produces a huge amount of data, a perception algorithm cannot be carried out in real-time without simplifying the sensor information. Unlike prior works, we introduce hybrid ground classification and the Region of Interest (ROI) identification method in order to filter out the amount of unwanted raw data for the actual tracking. And the environment is also abstracted based on an occupancy grid map. Moreover, we introduce feature based object geometry for precise estimation of the system state. In contrast to prior approaches, which use object geometry for the classification, we use it in order to compensate the unintended dynamics caused by shape change or occlusion. Our proposed MTT algorithm is able to run in real-time with an average processing time of 20ms. We evaluate it using our experimental vehicle “Leonie” in complex urban scenarios.

Defining and Substantiating the Terms Scene, Situation, and Scenario for Automated Driving

For the design and test of functional modules of an automated vehicle, it is essential to define interfaces. While interfaces on the perception side, like object lists, point clouds or occupancy grids, are to a certain degree settled already, they are quite vague in the consecutive steps of context modeling and in particular on the side of driving execution. The authors consider the scene as the central interface between perception and behavior planning & control. Within the behavior planning & control block, a situation is a central data container. A scenario is a common approach to substantiate test cases for functional modules and can be used to detail the functional description of a system. However, definitions of these terms are often-at best-vague or even contradictory. This paper will review these definitions and come up with a consistent definition for each term. Moreover, we present an example for the implementation of each of these interfaces.

Towards Tactical Lane Change Behavior Planning for Automated Vehicles

Recently, automated driving has more and more been transformed from an exciting vision into hands on reality by prototypes. While drivers are used to assistance and maybe even automation for driving within a lane, it is exciting to dare a step ahead: Deciding and executing tactical maneuvers like lane changes in automated vehicles without any human interaction. In this paper, we present our approach for tactical behavior planning for lane changes. We present a way to tackle perception uncertainties and how to achieve provident, prediction-based behavior planning. For this, we introduce a novel framework to plan in high-dimensional, mixed-integer state spaces in real-time. Our approach is evaluated not only in simulation, but also in real traffic. The implementation has recently been demonstrated to the public in the Audi A7 piloted driving concept vehicle, driving from Stanford to the Consumer Electronics Show (CES) 2015 in Las Vegas.

Ability and skill graphs for system modeling, online monitoring, and decision support for vehicle guidance systems

In this paper, the ability and skill graphs are introduced for modeling vehicle guidance systems in the concept phase of the development process (abilities), for online monitoring of system operation (skills), and to support driving decisions (skill levels) of automated road vehicles and advanced driver assistance systems. Both graphs rely on a decomposition of the human driving task. An ability is the entirety of conditions which are necessary to provide a certain part of the driving task. The ability graph can be developed in parallel to the item definition according to the ISO 26262 standard in the concept phase of the development process and can be used for supporting further development steps. A skill is defined as an abstract representation of a part of the driving task including information about the skills current performance. The skill graph is used to monitor the current system performance during operation and skill levels are input to driving decisions. Abilities and skills cover all aspects of the driving task including environment and self perception, data processing, decision making, and behavior execution. During operation of the developed item, the skill graph is instantiated as a (distributed) software component to process online information for assessing current skill levels. Each skill uses one or more performance metrics, which represent its current performance capability in relation to the maximum (inherent) ability level. The resulting information could replace the monitoring of the system by a human driver and can be used as an input to driving decisions of the vehicle to support appropriate and safe decisions.

Graph-based context representation, environment modeling and information aggregation for automated driving

The Stadtpilot project aims at fully automated driving on Braunschweigs inner city ring road. The TU Braunschweigs research vehicle Leonie is one of the first vehicles having the ability of fully automated driving in real urban traffic scenarios. In this paper, we present our approaches for context representation and environment modeling for automated driving. The demonstrated approach allows to provide a simple and yet universal information storage layer for the development of complex driving applications. Moreover, we present our approach for aggregating and fusing information between dynamic traffic objects detected by the sensor systems and a-priori map information.

Situation Assessment in Tactical Lane Change Behavior Planning for Automated Vehicles

Automated driving within a lane is a fascinating experience already. However, more exiting but also technically more challenging is to dare the next step of automating tactical behavior decisions for lane changes, as well. In this paper, we present our approach for situation assessment in tactical behavior planning for lane changes, whether lane changes are beneficial and/or possible. We present a way to tackle perception uncertainties and how to monitor the systems abilities and current skills. This is achieved by a dynamic Bayesian network and an unscented variance transform. Our approach is evaluated not only in a simulation, but also in real traffic. Our implementation has recently been demonstrated to the public in the Audi A7 piloted driving concept vehicle, driving 550 miles from Stanford to Las Vegas to the Consumer Electronics Show (CES) 2015.

Testing and Validating Tactical Lane Change Behavior Planning for Automated Driving

During the last 25 years, the driving abilities of automated vehicles have progressed rapidly. This went along with a huge increase of complexity for automated vehicles, regarding the multiplicity of interacting components being required to implement the functionality of automated vehicles.