Dirk M. Reichardt

CarTALK 2000: safe and comfortable driving based upon inter-vehicle-communication

CarTALK 2000 is a European Project focussing on new driver assistance systems which are based upon inter-vehicle communication. The main objectives are the development of cooperative driver assistance systems on the one hand and the development of a self-organising ad-hoc radio network as a communication basis with the aim of preparing a future standard. As for the assistance system, the main issues are: a) assessment of todays and future applications for co-operative driver assistance systems, b) development of software structures and algorithms, i.e. new fusion techniques, c) testing and demonstrating assistance functions in probe vehicles in real or reconstructed traffic scenarios. To achieve a suitable communication system, algorithms for radio ad-hoc networks with extremely high dynamic network topologies are developed and prototypes tested in the vehicles. Apart from technological goals, CarTALK 2000 actively addresses market introduction strategies including cost/benefit analyses and legal aspects, and aims at the standardisation to bring these systems to the European market. CarTALK 2000 started in August 2001 as a three-years project which is funded within the IST Cluster of the 5th Framework Program of the European Commission.

Motivation-based approach to behavior prediction

The design of advanced driver assistance systems always aims at enabling the driver to master todays traffic in a more safe and comfortable way. In order to judge the risks in a situation, the assistance system must be able to predict traffic behavior. Taking into account all possible future situations for the next few seconds is a task which quickly produces a complexity that can hardly be handled. Taking the human driver as a role model for its software-counterpart, we propose a new concept which aims at modeling anticipation by taking the motivations of drivers as a basis. Starting with a set of motivations typical for highway traffic, concrete situation specific goals are derived. A planning component generates the possible and fulfillable plans for all vehicles with respect to the goals. Then, the observed actions of the vehicles around are compared to these plans in order to derive a plausibility for the underlying intentions. Eventually, prediction is performed for plausible behaviors of vehicles, which are always based on a motivation that can be taken as an explanation for it. First results are shown in simulation for highway exit scenarios.

Shape Classification for Traffic Sign Recognition

Abstract A traffic sign detection and recognition approach is presented in this paper. This project is a part of the European research project PROMETHEUS(PROgraM for a European Traffic with Highest Efficiency and Unprecedented Safety) and is being developed by DAIMLER BENZ in collaboration with various university labs. Intensity segmentation, shape and traffic sign recognition have been joined together in a processing chain. Uncertainty handling, combining and propagation using Dernpster-Shafer rules fonn the heart of the shape recognition method. Multiple Knowledge Sources extract infonnation from the segmented image and increase knowledge about undefined shapes. Recognized shapes are transmitted to a high-level processing stage which perfonns model-based traffic sign recognition.

A real-time approach to traffic situation representation from image processing data

This paper discusses an approach to provide an environment representation of real traffic scenarios. Image processing is used to get information about the vehicles position on the road, about detected obstacles in the vicinity and about traffic signs. This incoming information is generated by camera modules which are arranged in a way that a panoramic vision is achieved. The problem is the noisy input data from different image processing modules which makes automated driving and adequate driver warning systems very difficult. The main task to solve this problem is to build up an always consistent situation representation in real time. Situation analysis basically consists of data fusion of different object recognition modules. An object matching algorithm has been developed that takes sensor specific deviations and errors into account. The second part is a rule set that examines the current situation for temporal or spatial inconsistencies using uncertainty representation. The resulting situation description is filtered by a set of Kalman filters which also provide a single step prediction. Apart from the fusion of obstacle recognition data, the situation analysis provides information about the road and currently valid traffic signs.

Approaching driver models which integrate models of emotion and risk

This paper discusses driver models as components of driver assistance systems and driving simulations. The focus is on generating believable and consistent behavior of simulated drivers by using driver models which imitate emotional influence on the human driverpsilas decisions. The designed and implemented simulation intends to build a platform for learning algorithms which will later be used in adaptive driver assistance systems. This work adapts the cognitive appraisal model as described by Orthony, Clore and Collins ill for the imitation of human emotional reactions by integrating it with a model of risk. A system design is presented and discussed. Moreover, a simulation environment is presented and the integration of emotional drivers is shown. The application of the model to a specific driver assistance system is presented.

A Definition Approach for an Emotional Turing Test

There are lots of modelling approaches for emotional agents. Can they be compared in any way? The intention of this work is to provide a basis for comparison in a small but consistent environment which focuses on the impact of emotions in the decision making of agents. We chose the public goods game with punishment option as a scenario. Why? In this scenario it has been proven that humans show emotional, non-rationalreactions. An emotional agent should therefore be able to show the same emotions and the underlying models should be capable of explainingthem! The simulation and test environment is designed to allow any emotional agent model. Eventually, human players should not be distinguishable from artificial emotional agents.

Exploring interactive teaching of a multi-modal emotional expression of a humanoid robot

One of the most popular domains in the current research is getting robots to replace humans in the difficult and dangerous tasks. To do that a robot must act as intelligently as humans towards different problems. Giving life to a robot is a dream that is coming true. One Of the hottest fields in this domain is teaching the robot how to understand and express emotions. The main target of this paper is to present an approach to teach a humanoid how to understand and then to express emotions. The robot should be able to understand emotions from plain text, voice and facial expressions then express them using gestures and voices. To validate the results, the robot should be able to read a text, understand the emotions in this text and then express them. Text synthesis is implemented through machine learning while expressing emotions is done through analyzing and then imitating peoples voices and postures while they experience different emotions. This research only focuses on the basic 3 emotions: happiness, sadness and anger. For the results an accuracy of 73.9% was reached in text synthesis, and survey was done to test the clarity of emotions expressed by the robot. An accuracy of 83.6%, 87.6% and 86.3% was reached in expressing happiness, sadness and anger respectively.

Analyzing Hand Therapy Success in a Web-Based Therapy System

After an injury, hand therapy should help regaining the full functionality of the hand, wrist and fingers. The application we developed helps a therapist in tracking the success of the therapy. Using the Leap Motion device as a basis, we calculate the individual finger joint angles. A second subsystem measures the angles between the fingers. This data is used to derive a report on the fingers‘ adduction and abduction during the therapy. The third subsystem detects the wrist rotation angles. Three algorithms were designed to detect the patient hand data, which the system feeds it to the three subsystems, the final algorithm max average error is 5.6 degrees, while the average max error is 9.3 degrees. The web-based system is part of a therapy support system, which allows patients to do their exercise at home and record their individual success.

Towards a Human Machine Interface Concept for Performance Improvement of Cycling

Exercising tends to be tedious and boring. Users search for means of entertainment during their workout. Motivation is an essential factor that inspires a person to maintain physical activity levels. A number of exercise bicycle manufacturers have enabled their equipment to port data signals to computers for the purpose of training logs. However, they all lack the motivational and the fun factors in their system. Many people face obstacles to improve their physical activity due to lack of accessibility to physical activity facilities and physical trainers. Technology can provide immediate personal assistance anywhere and anytime. In this paper, we present a persuasive game environment integrated to cycle training. The system measures the performance of a rider and visualizes a scenario through which the virtual bike trip leads. The players maintain their excitement while riding the bicycle to improve their performance on how to enhance cycling skills.

Towards Virtual Emotions and Emergence of Social Behaviour

In our approach we explore emotional effects of virtual agents withouthuman interaction. The idea is the following: emotion is considered a valuable means to explain social phenomena like the emergence of costly punishment. Now, would it be possible to put together a population of agents which mimics human society? Would it even be possible to show that certain personality configurations and emotional behavior emergefrom interactions in a virtual agent society experiment? In previous work we selected the public goods game with punishment option because of its intriguing property of affect-based decisions [1]. In this paper we present our first approach to virtual agent population modelling which adopts a setting of the public goods game as it is used for studies on cooperative behaviour (see [1][3]). The discrete version of the game distinguishes between the roles of the punisher, the cooperator and the defector. As shown in [3] the iterated public goods game leads to a predominance of the defectors. This is changed by giving the participants the freedom to choose whether to participate in the game or not which gives the punishers a chance to get back into the game and it is shown that eventually they become dominant (see [3]).