The fight against climate change will among other measures require new mobility concepts. Promoting electric vehicles is one key challenge in order to reduce CO2 emissions which are produced to a significant amount by combustion engine cars. Traffic congestions in cities often originate from vehicles searching for a suitable parking spot.
The V-Charge project tackles these issues by introducing a new mobility concept mitigating the two main deficiencies of electric vehicles nowadays: reduced driving ranges and increased refuelling durations. Driverless parking and charging is promoted to ease the traveller’s transfer from individual short-range traffic with his or her electric vehicle to long haul public traffic by train or airplane. Relieving the traveller from the time-consuming task of searching for a parking spot will render public transportation more appealing. This will not only help to increase the attractiveness of electric vehicles, but additionally the one of more environmentally friendly public transportation.
This concept requires fully automated driving in indoor and outdoor parking lots. Consequently built around the vision to promote electric vehicles equipped with functionality to realize this mobility concept, the V-Charge project sets itself the challenging task to achieve such automated driving capabilities with close-to-market, low-cost sensors only. Keeping an eye on the additional price for the customer is mandatory to gain high customer acceptance. Utilizing a sensor setup only consisting of four monocular fisheye cameras, two stereo cameras and stock ultrasonic sensors pushes the price of the V-Charge system to a minimum, yet requires increased scientific and engineering efforts to process the sensor data obtained. As parking lot owners and operators are likewise among the potential customers of the V-Charge system, the project additionally focus on keeping expenses and overhead for those to a minimum. By relaxing constraints on modifications and maintenance efforts of the required infrastructure the project aims at developing a system offering substantial benefits not only to drivers but also to parking lot operators. Automatic release of potentially limited charging stations by the vehicles themselves as well as high density parking renders the V-Charge system attractive for those.
As the Vienna Convention on Road Traffic signed by numerous countries worldwide renders fully automated driving in public traffic difficult from a legal point of view, the V-Charge application domain in designated parking areas also constitutes a promising approach to pave the way for automated vehicles. Smaller legal hurdles on private property or closed-off areas will facilitate the introduction of automated vehicles in real-world applications. Customer acceptance and trust can successively be established by proving long-term reliable operation in this controlled domain.
From a scientific and engineering point of view the realization of the V-Charge concept requires progress in the state of the art in various research domains. Autonomous indoor navigation without modifications to infrastructure and environment requires GPS independent localization with onboard sensors only. Fully automated parking in tight spaces calls for precise environment perception and control of the vehicle. Operation in parking lots utilized by both automated and manual operated cars requires detection, classification and estimation of the intents of other traffic participants. And last but not least a convenient interface for the user is required making dropoff and pickup of the vehicle as easy as possible, despite the complex scheduling algorithms for parking spot and charging station assignment operating in the background.
|Workpackage 1 is concerned with the development of the vehicle platform and low-level perception system. During the fourth year an electric Golf VII platform was equipped and modified to support fully automated operation, complementing the two Golf VI combustion engine test vehicles. Each vehicle is equipped with a surround-view camera system, front and rear looking stereo cameras, and stock vehicle hardware such as ultrasonic sensors and wheel odometers, as well as actuation of the steering wheel, throttle, and brake. An inductive charging demonstrator was developed enabling the V-Charge team to demonstrate the project´s vision from start to end.|
|Work package two focuses on creating and maintaining geometric and semantic maps of operating areas. The project has developed a topological metric mapping concept, which contains layered information about an area’s appearance, geometry, and high-level structure (such as the road network, and the number and position of the parking spots and charging areas). The fourth year saw the expansion of the map’s semantic layer including automatic parking spot segmentation and speed scheduling so slow down when the car approaches areas where many pedestrians have been observed.|
|The third work package addresses vision-based localization and online perception for static and dynamic objects. In year four, the visual localization system was thoroughly tested, achieving robust long-term localization results over varying environmental conditions with an accuracy of smaller than ten centimetres. The capabilities of the vehicle were extended to measure the environment and to detect, label and track objects therein directly from the monocular fisheye camera images.|
|The goal of work package four is to develop motion planning and vehicle control algorithms suitable for automated driving in the presence of other vehicles and pedestrians. In the project’s fourth year, we robustified platooning in the presence of other vehicles and incorporated safe navigation among pedestrians and at intersections. Parking accuracy was improved even further, enabling precise manoeuvres to perfectly position the vehicle on inductive charging plates.|
|The objective of work package five is to develop a back-end server and communication architecture to serve as the infrastructure management system. The fourth year of the project saw a redesign of the android application based on a design study as well as extensive evaluation of the communication and security architecture regarding the scalability to handle large-scale parking lots with up to thousands of parking spots.|
|Work package six is concerned with testing, system integration, and demonstration. In June and July 2015, the consortium had multiple successful public demonstrations at the Mobile Life Campus in Wolfsburg and the Excellence Parking Garage at Schiphol Airport, Amsterdam. During the live demonstrations, all components involved in driverless parking and charging such as camera-only localization, obstacle avoidance, reaction to pedestrians and vehicles in mixed-traffic scenarios and smartphone-triggered dropoff and pickup of the vehicle were shown to both technical experts and international media. These efforts resulted in numerous media entries on prestigious worldwide platforms.|
The goal of work package seven is to ensure the high visibility of the V-Charge project through dissemination and exploitation of results. The consortium aims to develop novel scientific contributions for all aspects of the project. To this end, we have published 56 scientific papers throughout the course of the project in international conferences, workshops, and journals and released 3 major open-source code projects.
The eighth work package covers management of the project.
In the course of the project’s four years, three fully functional prototypes capable of automated operation in both indoor and outdoor parking lots and garages were developed. Two combustion engine Golf VI platform and one fully electric Golf VII prototypes were equipped with close-to-market camera sensors and navigation software developed in the project. The prototypes were successfully demonstrated at various locations including several indoor parking garages as well as one medium-scale outdoor parking area.
A typical V-Charge mission begins with the dropoff of the vehicle by the customer in a designated dropoff area and the required clearance confirmation via his or her smartphone to send the vehicle on its way to a parking spot or charging station. In this context, the project developed an android application for this purpose involving a user design study to optimize the usability. The project additionally developed secure data transmission concepts between vehicle(s) and a remote parking lot server accessed via local wireless links or mobile data connections. Handling of large amounts of parking spots and charging stations by the parking lot server were demonstrated in simulation.
After having received the user’s clearance approval, the vehicle localizes itself in the environment. The localization task is tackled with close-to-market monocular cameras and natural landmarks only to strive for cost efficiency for both vehicle owners and parking lot operators. During localization, the vehicle compares perceived images to an offline map database containing visual information about the parking lot. This localization technique was optimized throughout the course of the project to yield centimeter-level accuracy and to deal with a large variety of lighting conditions and environmental changes. To achieve this level of accuracy, superior calibration of the four monocular surrounding the car is required. The V-Charge project members developed software to calibrate and work with such a novel fisheye multi-camera setup and released with “OpenGV” and “CamOdoCal” two publicly available software suites.
Being localized and sent on its way, the V-Charge vehicle starts to follow a route planned on a topological map of the parking lot enhanced with semantic information like parking spot locations and speed limits derived from offline mapping runs in a semi-supervised fashion. Introspective classification is used to reason about the uncertainty of labels that the algorithm assigns to objects, asking for human expertise where the software-based classifier is too uncertain. This approach allows to incrementally incorporate human expertise and enables the classifier to improve over time requiring less and less feedback from the human expert resulting in a transferable and low-maintenance mapping process.
While navigating along the route, the vehicle perceives its surrounding and separates static obstacles from other mobile traffic participants. Pedestrians and other vehicles are detected and classified from the monocular cameras 360° degrees around the vehicle. Accurate depth information is extracted from the monocular cameras to complement the front and rear facing stereo cameras with static obstacle information to the sides, especially useful to measure parking spot widths.
A local motion planner computes safe motion commands for the platform, safely mitigating static obstacles and negotiating with other pedestrians and vehicles in mixed traffic. Here the project developed an approach complementing the state of the art in motion planning for autonomous vehicle.
Having arrived at the designated parking spot or charging station, the V-Charge vehicle monitors the assigned parking spot to determine its occupancy state in case a manual driven vehicle parked in the very same spot. For this task the side monocular cameras are used. In this context, the V-Charge project pushed the state of the art in dense motion stereo from monocular cameras. The vehicle then computes a potentially complex manoeuvre into the parking spot/charging bay. Parking precision was pushed to a few centimetres, certainly exceeding human skills.
The project yielded valuable insights in the background scheduling process of parking spots and charging stations including concepts for high density parking via numerous publications in this domain.