In this paper we present algorithms for automatically generating a road network description from aerial imagery. The road network inforamtion (RNI) produced by our algorithm includes a composite topoloigical and spatial representation of the roads visible in an aerial image. We generate this data for use by autonomous vehicles operating on-road in urban environments. This information is used by the vehicles to both route plan and determine appropriate tactical behaviors. RNI can provide important contextual cues that influence driving behaviors, such as the curvature of the road ahead, the location of traffic signals, or pedestrian dense areas. The value of RNI was demonstrated compellingly in the DARPA Urban Challenge 1, where the vehicles relied on this information to drive quickly, safely and efficiently. The current best methods for generating RNI are manual, labor intensive and error prone. Automation of this process could thus provide an important capability. As a step toward this goal, we present algorithms that automatically build the skeleton of drivable regions in a parking lot from a single orthoimage. As a first step in extracting structure, our algorithm detects the parking spots visible in an image. It then combines this information with the detected parking lot boundary and information from other detected 1 The Urban Challenge (or the DARPA Urban Challenge) was an autonomous vehicle competition in which competitors had to build vehicles capable of auotnomously driving 60 miles amongst moving traffic in an urban environment. Visit

to develop autonomous vehicles capable of safely, reliably, and robustly driving in traffic. In this article we introduce Boss, the autonomous vehicle that won the challenge. Boss is a complex artificially intelligent software system embodied in a 2007 Chevy Tahoe. To navigate safely, the vehicle builds a model of the world around it in real time. This model is used to generate safe routes and motion plans both on roads and in unstructured zones. An essential part of Boss’s success stems from its ability to safely handle both abnormal situations and system glitches. In 2003 the Defense Advanced Research Projects Agency (DARPA) announced the first Grand Challenge with the goal of

Abstract — In this paper we describe a state lattice based motion planning approach, which we have successfully applied to large, cluttered, but quasi-static environments. Our approach produces smooth and complex maneuvers through the use of a multi-resolution state lattice, where the resolution is adapted based on the environment, and distance from the robot. We also describe a framework for detecting dynamic obstacles such as pedestrians and cars using a multisensor lasercamera detection and tracking method. Image detection is based on several extensions to the Implicit Shape Model technique; laser detection is instead achieved through the use of a Conditional Random Fields reasoning. Objects are tracked through the use of multiple motion model Kalman filters in order to cope with several different motion dynamics. Urban environments, are complex, cluttered, and dynamic scenes, however. We therefore propose to extend our dynamic obstacle detection and tracking method with a short-term motion prediction functionality based on the same models used for tracking, effectively generating time based cost or risk maps. We further propose to implement these cost maps into our high-dimensional (5D to 6D) lattice planner to generate time-optimal trajectories in dynamic, cluttered environments. A D * implementation is envisioned to speed up re-planning dramatically. I.

detect and correct large discrepancies between the heuristic cost-to-go estimate and the true cost function. In situations where these large discrepancies exist, the search may expend significant effort escaping from the “bowl ” of a local minimum. A * typically computes supporting data structures for the heuristic once, prior to initiating the search. FAHR directs the search out of the bowl by recomputing parts of the heuristic function opportunistically as the search space is explored. FAHR may be used when the heuristic function is in the form of a pattern database. We demonstrate the effectiveness of the algorithm through experiments on a ground vehicle path planning simulation. I.

For decades, humans have dreamed of making cars that could drive themselves, so that travel would be less taxing, and the roads safer for everyone. Toward this goal, we have made strides in motion planning algorithms for autonomous cars, using a powerful new computing tool, the parallel graphics processing unit (GPU). We propose a novel five-dimensional search space formulation that includes both spatial and temporal dimensions, and respects the kinematic and dynamic constraints on a typical automobile. With this formulation, the search space grows linearly with the length of the path, compared to the exponential growth of other methods. We also propose a parallel search algorithm, using the GPU to tackle the curse of dimensionality directly and increase the number of plans that can be evaluated by an order of magnitude compared to a CPU implementation. With this larger capacity, we can evaluate a dense sampling of plans combining lateral swerves and accelerations that represent a range of effective responses to more on-road driving scenarios than have previously been addressed in the literature. We contribute a cost function that evaluates many aspects of each candidate

This paper focuses on the development of a follow behavior for an unmanned ground vehicle (UGV) in collaborative scenarios. The scenario being studied involves a human traveling over a rugged terrain on foot. The UGV follows the human. We present an approach for automatically generating a reactive energy-efficient follow behavior that maps the vehicle’s states into motion goals. We start by partitioning the state space that encodes the relationship between the state of the vehicle and the human’s state, and the environment. For each cell in the partitioned state space, we either directly generate the motion goal for the vehicle to execute or a function that produces the motion goal. The motion goal defines not only the location towards which the vehicle should move but also specifies a zero activity zone around the human within which the vehicle is supposed to slow down and remain stationary to save its energy until it gets outside the margin caused by the movement of the human. Our approach utilizes off-line simulations to assess the performance of the generated behavior. Our simulation results show that the automatically generated follow behavior significantly outperforms a simple conservative tracking rule in terms of distance traveled and violation of proximity constraints. We anticipate that the approach presented in this paper will ultimately enable us to implement energy efficient follow behaviors on physical UGVs. 1

This research was supported through the General Motors/Carnegie Mellon