Back to search

Democritus University of Thrace (DUTH)

Νέες τεχνικές χάραξης πορείας ρομπότ και ρομποτικής όρασης

Abstract

dc:description

The research field of robotics is one of the most wide-spread and vibrant scientific area. Different types of robots are used to fulfill numerous tasks while they are equipped with a variety of sensors. The appropriate usage of the available resources and sensors of a robotic system constitutes a crucial factor for retrieving valuable information related to the status of the surrounding environment. The sensors are the means through which the robot can perceive and interact with its environment. The applications of the robotic systems as a whole consider the exploitation of the sensors’ readings as a major prerequisite to accomplish specific tasks.The techniques of creating collision free trajectories as well as robot vision techniques are the most well-known examples of methods and applications that require the use of sensors. The main objective of this PhD thesis is the study and the development of new techniques for the creation of robot trajectories as well as robot vision approaches related to the navigation of mobile robots. The first described technique for trajectory creation is based on the application of Cellular Automata (CA) in order for a team of cooperative robots to follow a predefined trajectory. The main goal of the system is to provide a locomotion framework to a robotic team so that the desired distance could be covered while possible detected obstacles are avoided. In addition, the method for creating trajectories must exploit and utilize the minimum potential resources in order to fulfill further goals such as formation control of the team via processes of cooperation. Initially, the configuration space is divided into a number of identical square cells which are characterized by a single state at each time step. The state of each cell is evolved at each time step according to its current state and the corresponding states of its adjacent cells by applying the appropriate transition rules. The desired distance, the collision avoidance and the formation control processes are completed with the application of suitable CA transition rules. Due to the CA simplicity, the proposed technique displays low technical requirements for its implementation in real systems. Dynamically moving objects can be detected and avoided successfully, though with some restrictions due to the discretization of space. Compared to other similar techniques reported in the literature which have high requirements concerning the characteristics of the robots in use, the proposed method could be implemented using a robot team consisting of robots with limited technical specifications.The basic problem that must be confronted in multitudinous robot teams relies on the fact that local cooperation among numerous adjacent robots is required in order to preserve their initial formation. Despite its importance, this cooperation process proves to be time consuming since the formation recovery is achieved at local level. The second technique for trajectory creation was proposed in order to minimize the required time for cooperation and simplify the formation maintenance process. The proposed technique is based on the synergy of two fundamental artificial intelligence (AI) methods, the CA and the Artificial Ant Colonies (AAC) algorithms. The basic tool that was exploited from the AAC algorithms was the substance “pheromone” in order to minimize the required processes of communication between the adjacent robots to the absolutely necessary ones. Initially, the entire robot team is divided into equally numbered subgroups which must form and preserve the same pattern. The configuration area is analyzed in a lattice of identical cells which are denoted with specific states at each time step. The states characterize cells which correspond to: free space, a robot of the team, an object of the space and a certain amount of pheromone. The state of each cell is evolved according to the application of the transition CA rules which simulate all the required functions of the system. The formation control operation is achieved either through the direct communication of the adjacent robots or the detection of a “pheromone” quantity. During its route to its final position, each robot imprints its course on the ground with the use of a substance which is characterized by the main feature of the pheromone, the evaporation. The “pheromone” quantity is discretized similarly to the free configuration space so that it can be denoted as CA state and detected by the robots of the forthcoming subgroups via the appropriate sensors. Consequently, the cooperation processes are not executed consecutively since possible formation modifications are denoted on the ground by precedent subgroups. The proposed technique was implemented with the use of a real world and systems simulator, Webots, mainly due to its embedded components which can simulate the “pheromone” functions. The used robots were equipped with additional obstacle detection sensors so that the negative effects of the area discretization could be minimized. Additionally, due to the CA inherited parallelism, all the CA cells evolve their state simultaneously and thus, the technique displays low execution times compared to other related approaches. Finally, the required resources are minimized due to the simplicity of both the AI techniques compared to other trajectory creation methods which require complex and higher cost robots.The creation of collision free trajectories constitutes a significant research area of robotics since the navigation ability is provided to a robot. The robot navigation and the retrieval of optical data of a captured scene are two directly related fields. Area mapping and surveillance with the use of mobile robots are two fundamental examples of such applications. The techniques which retrieve and process data acquired from digital cameras belong in the broader research field of robot vision. For this reason, a novel robot vision technique was developed with which an appropriate record of the configuration area is formed and can be subsumed in the general field of the digital image stabilization methods. The image stabilization process compensates the unwanted motion of a camera which occurs during the image sequence acquisition. The proposed technique is based on the application of the Hilbert-Huang transformation (HHT), a signal analysis method. The total observed motion of image sequences which were acquired by a mobile camera is mainly caused by two different motions: the desired camera motion and the undesired jitter motion which is produced by the trembled motion of the camera. Local motion vectors are initially estimated from specific image areas of the background while the resulted signal is decomposed into a finite number of subsignals with the application of the Empirical Mode Decomposition (EMD) process. Subsequently, the energy content of each subsignal is extracted with the application of the Hilbert transform in order to define the last subsignal which will include components of the undesired camera motion. The summation of the initially produced subsignals corresponds to the jitter camera motion and, thus, it is removed from the original image sequence so that the stabilized version could be created. Experimental results have shown that the proposed technique exhibits superior performance compared to other digital image stabilization algorithms. The qualitative superiority of the method over other reported approaches relies on the fact that the analysis of the initial motion signal is based on the basic features of each independent motion. The undesired jitter motion includes higher frequency components with lower pixel displacements while the components with contrary features comprise the desired camera motion. A software scheme was developed which cooperates with a real mobile robot, executing all the required navigation processes and producing the stabilized versions of the acquired image sequences with the proposed technique.Finally, a novel technique related to the adaptable retrieval of the third dimension of space is proposed. The extracted disparity maps are usually utilized in applications such as autonomous mobile robot navigation and unknown terrain mapping. The proposed technique recalculates the value of a significant variable in depth extraction, the disparity level, which essentially corresponds to the maximum pixel value of the extracted disparity map. A depth map is initially calculated with the application of any stereo correspondence approach and, thus, the corresponding histogram displays a number of possible disparities equal to the disparity level. Sequentially, the resulted histogram is analyzed into equally sized parts and the disparity level redefinition is achieved with the calculation of statistical metrics for the right-most histogram part. In case of an adjacent object, a distribution increment appears on the last subhistogram and, thus, a disparity level increment is required in order to continuously extract the required disparity values. On the contrary, a high disparity level value is redundant and results to a higher computational burden. In order to reduce the complexity of the stereo correspondence approach, the disparity level is decreased. All the reported stereo correspondence methods consider fixed values for the disparity map extraction which actually excludes the dynamic features of the natural scenes. The application of the proposed robot vision technique could accommodate and improve the navigation of a mobile robot since the reduction of the disparity level, which eventually leads to higher frame rates, is possible, much like the detection of adjacent objects. A software framework was developed which acquires stereo image pairs from a real robot and executes all the required processes of the proposed stereo vision method. The required processing time for the disparity level redefinition is kept at low levels rendering the proposed method appropriate for many robot applications.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2013

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Ιωαννίδης, Κωνσταντίνος
  • Ioannidis, Konstantinos

Subjects

dc:subject × 20

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/30497

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Ιωαννίδης, Κωνσταντίνος; Ioannidis, Konstantinos. Νέες τεχνικές χάραξης πορείας ρομπότ και ρομποτικής όρασης. Democritus University of Thrace (DUTH), 2013. http://hdl.handle.net/10442/hedi/30497