Paul Levi

3D object recognition from range images using local feature histograms

The paper explores a view-based approach to recognize free-form objects in range images. We are using a set of local features that are easy to calculate and robust to partial occlusions. By combining those features in a multidimensional histogram, we can obtain highly discriminant classifiers without the need for segmentation. Recognition is performed using either histogram matching or a probabilistic recognition algorithm. We compare the performance of both methods in the presence of occlusions and test the system on a database of almost 2000 full-sphere views of 30 free-form objects. The system achieves a recognition accuracy above 93% on ideal images, and of 89% with 20% occlusion.

Locating Biologically Active Compounds in Medium-Sized Heterogeneous Datasets by Topological Autocorrelation Vectors: Dopamine and Benzodiazepine Agonists

Electronic properties located on the atoms of a molecule such as partial atomic charges as well as electronegativity and polarizability values are encoded by an autocorrelation vector accounting for the constitution of a molecule. This encoding procedure is able to distinguish between compounds being dopamine agonists and those being benzodiazepine receptor agonists even after projection into a two-dimensional self-organizing network. The two types of compounds can still be distinguished if they are buried in a dataset of 8323 compounds of a chemical supplier catalog comprising a wide structural variety. The maps obtained by this sequence of events, calculation of empirical physicochemical effects, encoding in a topological autocorrelation vector, and projection by a self-organizing neural network, can thus be used for searching for structural similarity, and, in particular, for finding new lead structures with biological activity.

Advanced lane recognition-fusing vision and radar

One major problem of the common vision-based lane recognition systems is their susceptibility to weather. These problems mainly stem from the fact, that they only look for road structures. From the position of other cars in front, the run of the curve can be estimated. This paper presents our fusion approach, that takes leading vehicles into account which have been detected by radar. The Kalman filter applied here does not only deliver improved measurements of the run of the curve, but also a precise estimate of the lateral position of the observed cars. This information can be used to improve the lane assignment of ACC systems.

Minimalistic approach towards communication and perception in microrobotic swarms

This work is primarily devoted to specific communication and sensing approaches applied for large microrobotic swarms. We investigate the minimal capabilities of a microrobot which still enable the whole robotic group to perform collective activities. These minimal capabilities are implemented in the hardware which allows exploring a phenomenon of swarm intelligence in real experiments. The components of the developed system consume energy provided by microcontrollers I/O ports, are cheap and available on micro-component market.

Cooperative Multi-Robot Path Planning by Heuristic Priority Adjustment

In this paper, a new algorithm for cooperative path planning in a multi-robot system is introduced. The algorithm is specially designed for a distributed system with a large number of robots, all of which should be able to reach their respective goal positions without blocking each other, even if the environment is heavily constricted by obstacles. The basic idea of the approach is based on fully distributed path planning without any central instance, but with the ability of communication and cooperation between the robots. Path planning is done on local sections of the time-space configuration space. Dynamic conflicts between the robots are solved by the heuristic adjustment of priority values. By a continuous enhancement of all plans the algorithm is very robust against dynamic changes and erroneous robot behavior

From real robot swarm to evolutionary multi-robot organism

Collective working allows microrobots to achieve more functionality, better performance and higher reliability on the macroscopic level. In this paper we demonstrate the on-going work in developing novel collective systems, where swarm robots work not only collectively, but are also capable of autonomous aggregation and disaggregation into a higher multi- robot organism. The main issues of such an organism, as well as its genome-based control, are discussed. We show the developed docking approach and investigate topological transformations in a prototype of self-assembling robots.

A new approach to exploiting parallelism in ant colony optimization

In the paper, a new approach to exploiting parallelism in ant colony optimization (ACO) is implemented on a supercomputer (Cray T3E). Unlike the previous methods where results were based on either simulation or independent executions, in this paper based on the implementation we have studied the issues of parallelization and the overhead of communications apart from the idle times required in case of synchronous communication. The results are compared with already available methods. Moreover, by varying the values of different parameters, the effects are also analyzed for this method. Albeit the optimization method being general, TSP (Traveling Salesman Problem) is chosen for experimentation as it is widely researched and standard benchmarks are also available. The communication interval balances the total communication time and the frequency of global update. At the same time the best ant in each colony alone is allowed to update globally, even though locally all ants update the pheromone trails. The results obtained convince the efficiency of the approach.

APPLICATION OF ORDER PARAMETER EQUATIONS FOR THE ANALYSIS AND THE CONTROL OF NONLINEAR TIME DISCRETE DYNAMICAL SYSTEMS

This work is based on the concept of order parameters of synergetics. The order parameter equations describe the behavior of a system in the vicinity of an instability and are used here not only for the analysis but also for the control of nonlinear time discrete dynamical systems. Usually, the dimensionality of the evolution equations of the order parameters is less than the dimensionality of the original evolution equations. It is, therefore, convenient to introduce control mechanisms, first in the order parameter equations, and then to use the obtained results for the control of the original system. The aim of the control in this case is to avoid chaotic behavior of the system. This is achieved by shifting appropriate bifurcation points of a period-doubling cascade. In this work we concentrate on the shifting of only the first bifurcation point. The used control mechanisms are delayed feedback schemes. As an example the well-known Henon map is investigated. The order parameter equation is calculated using both the adiabatic elimination procedure and the center manifold theory. Using the order parameter concept two types of control mechanisms are constructed, analyzed and compared.

Stability of On-Line and On-Board Evolving of Adaptive Collective Behavior

This work focuses on evolving purposeful collective behavior in a swarm of Jasmine micro-robots. We investigate the stability of the on-line and on-board evolutionary approaches, where mutation, crossover as well as fitness calculation are performed only by interacting micro-robots without using any centralized resources. In this work it is demonstrated that the environment-adaptive collective behavior can be obtained, where the evolving fitness and behavior are partially stable. To increase stability of the approach, some reduction methodology of the search space is proposed.

Detecting reflection posts - lane recognition on country roads

This paper presents a new approach to the challenging task of lane recognition on general roads. Lane recognition is the basis for many driver assistance systems, including lane departure warning and the assignment of vehicles to specific lanes. Most systems of the past are designed for the well defined highway scenario. They rely on white lane markings with known geometric appearance. Our method is an extension which also works when lane markings are in bad conditions or missing completely. We use reflection posts as a means of estimating the course of the road. In addition we are furthermore able to measure the horizontal and vertical slope of the road surface.