Tesi di dottorato >
Ingegneria industriale e dell'informazione >
Please use this identifier to cite or link to this item:
|Title: ||Sviluppo di metodi e apparati per la navigazione del robot e per la mappatura simultanea di ambienti non strutturati.|
|Other Titles: ||Methods and Devices for Mobile Robot Navigation and Mapping in Unstructured Environments|
|Authors: ||Mumolo, Enzo|
real time operating system
|Issue Date: ||18-Mar-2008|
|Publisher: ||Università degli studi di Trieste|
|Abstract: ||The work described in this thesis has been carried out in the context of the exploration of an unknown environment by an autonomous mobile robot. It is rather difficult to imagine a robot that is truly autonomous without being capable of acquiring a model of its environment. This model can be built by the robot exploring the environment and registering the data collected with the sensors over time.
In the last decades a lot of progress has been made regarding techniques focused on environments which posses a lot of structure. This thesis contributes to the goal of extending existing techniques to unstructured environments by proposing new methods and devices for mapping in real-time.
The first part of the thesis addresses some of the problems of ultrasonic sensors which are widely used in mobile robotics for mapping and obstacle detection during exploration. Ultrasonic sensors have two main shortcomings leading to disappointing performance: uncertainty in target location and multiple reflections. The former is caused by wide beam width and the latter gives erroneous distance measurements because of the insertion of spikes not directly connected to the target.
With the aim of registering a detailed contour of the environment surrounding the robot, a sensing device was developed by focusing the ultrasonic beam of the most common ultrasonic sensor to extend its range and improve the spatial resolution. Extended range makes this sensor much more suitable for mapping of outdoor environments which are typically larger. Improved spatial resolution enables the usage of recent laser scan matching techniques on the sonar scans of the environment collected with the sensor. Furthermore, an algorithm is proposed to mitigate some undesirable effects and problems of the ultrasonic sensor. The method registers the acquired raw ultrasonic signal in order to obtain a reliable mapping of the environment.
A single sonar measurement consists of a number of pulses reflected by an obstacle. From a series of sensor readings at different sonar angles the sequence of pulses reflected by the environment changes according to the distance between the sensor and the environment. This results in an image of sonar reflections that can be built by representing the reading angle on the horizontal axis and the echoes acquired by the sensor on the vertical one. The characteristics of a sonar emission result in a texture embedded in the image. The algorithm performs a 2D texture analysis of the sonar reflections image in such a way that the texture continuity is analyzed at the overall image scale, thus enabling the correction of the texture continuity by restoring weak or missing reflections.
The first part of the algorithm extracts geometric semantic attributes from the image in order to enhance and correct the signal. The second part of the algorithm applies heuristic rules to find the leading pulse of the echo and to estimate the obstacle location in points where otherwise it would not be possible due to noise or lack of signal. The method overcomes inherent problems of ultrasonic sensing in case of high irregularities and missing reflections.
It is suitable for map building during mobile robot exploration missions. It's main limitation is small coverage area. This area however increases during exploration as more scans are processed from different positions.
Localization and mapping problems were addressed in the second part of the thesis. The main issue in robot self-localization is how to match sensed data, acquired with devices such as laser range finders or ultrasonic range sensors, against reference map information. In particular scan matching techniques are used to correct the accumulated positional error using dead reckoning sensors like odometry and inertial sensors and thus cancel out the effects of noise on localization and mapping.
Given the reference scan from a known position and the new scan in unknown or approximately known position, the scan matching algorithm should provide a position estimate which is close to the true robot position from which the new scan was acquired.
A genetic based optimization algorithm that solves this problem called GLASM is proposed. It uses a novel fitness function which is based on a look up table requiring little memory to speed the search. Instead of searching for corresponding point pairs and then computing the mean of the distances between them, as in other algorithms, the fitness is directly evaluated by matching points which, after the projection on the same coordinate frame, fall in the search window around the previous scan. It has a linear computational complexity, whereas the algorithms based on correspondences have a quadratic cost.
The GLASM algorithm has been compared to it's closest rivals. The results of comparison are reported in the thesis and show, to summarize, that GLASM outperforms them both in speed and in matching success ratio.
Glasm is suitable for implementation in feature-poor environments and robust to high sensor noise, as is the case with the sonar readings used in this thesis which are much noisier than laser scanners. The algorithm does not place a high computational burden on the processor, which is important for real world applications where the power consumption is a concern, and scales easily on multiprocessor systems. The algorithm does not require an initial position estimate and is suitable for unstructured environments.
In mobile robotics it is critical to evaluate the above mentioned methods and devices in real world applications on systems with limited power and computational resources. In the third part of the thesis some new theoretical results are derived concerning open problems in non-preemptive scheduling of periodic tasks on a uniprocessor.
This results are then used to propose a design methodology which is used in an application on a mobile robot. The mobile robot is equipped with an embedded system running a new real-time kernel called Yartek with a non-preemptive scheduler of periodic tasks. The application is described and some preliminary mapping results are presented.
The real-time operating system has been developed in a collaborative work for an embedded platform based on a Coldfire microcontroller. The operating system allows the creation and running of tasks and offers a dynamic management of a contiguous memory using a first-fit criterion. The tasks can be real-time periodic scheduled with non-preemptive EDF, or non real-time. In order to improve the usability of the system, a RAM-disk is included: it is actually an array defined in the main memory and managed using pointers, therefore its operation is very fast.
The goal was to realize small autonomous embedded system for implementing real-time algorithms for non visual robotic sensors, such as infrared, tactile, inertial devices or ultrasonic proximity sensors. The system provides the processing requested by non visual sensors without imposing a computation burden on the main processor of the robot. In particular, the embedded system described in this thesis provides the robot with the environmental map acquired with the ultrasonic sensors.
Yartek has low footprint and low overhead.
In order to compare Yartek with another operating system a porting of RTAI for Linux has been performed on the Avnet M5282EVB board and testing procedures were implemented. Tests regarding context switch time, jitter time and interrupt latency time are reported to describe the performance of Yartek.
The contributions of this thesis include the presentation of new algorithms and devices, their applications and also some theoretical results.
They are briefly summarized as:
A focused ultrasonic sensing device is developed and used in mapping applications.
An algorithm that processes the ultrasonic readings in order to develop a reliable map of the environment is presented.
A new genetic algorithm for scan matching called GLASM is proposed.
Schedulability conditions for non-preemptive scheduling in a hard real-time operating system are introduced and a design methodology is proposed.
A real-time kernel for embedded systems in mobile robotics is presented.
A practical robotic application is described and implementation details and trade-offs are explained.|
|Appears in Collections:||Ingegneria industriale e dell'informazione|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.