Ebook: Intelligent Autonomous Systems 11

Welcome to the l1th International Conference on Intelligent Autonomous Systems (IAS-11). Since 1986, the IAS conferences have been meeting places for researchers on Intelligent Systems The goal of the IAS conference is to bring together leading researchers interested in all aspects of autonomy and adaptivity of artificial systems. A current trend which is also reflected in this conference is the interaction with humans of intelligent systems where there is a symbiosis between humans and intelligent systems.

This year, 35 papers have been submitted to IAS-11. All the submissions were rigorously reviewed by the Program Committee. A total of 25 papers were accepted for presentation at the conference. This IAS-11 conference features a single track of technical presentations of papers with high scientific quality. The papers contained in the final program cover a wide spectrum of research in autonomous intelligent systems including interactive systems, learning, perception, localization and mapping, navigation planning and distributed systems.

The proceedings include all accepted papers and reflect a variety of topics concerning intelligent autonomous systems. The organizers would like to express their gratitude to all contributors in the preparation phase as well as during the meeting. We would especially like to thank the program committee members for their valuable support and for the preparation of the reviews, which allowed making a proper selection of high-quality papers.

The staff of the School of Information Technology and Engineering (SITE) of the University of Ottawa and especially Dr. Ana-Maria Cretu took a great part in planning and organizing the conference.

We wish all participants that they enjoy IAS-11 and the beautiful city of Ottawa. We hope that IAS-11 will provide you with new ideas, allow you to exchange knowledge, and be a prosperous event for you.

Henrik Christensen,

Frans Groen,

Emil Petriu

A nurse activity efficiency is controlled greatly by the order of activities, and the system for nurses to learn the ideal order of activities is expected. For the construction of this system, it is required to recognize nurse activities from the sensor data, to calculate the order of activities and to compare the activity order that is recognized with other nurses' or optimized activity order. In this paper, we used accelerometers and RFID tags to recognize these nursing activities. We verified the accuracy of activity recognition and determined the activities that could not be easily recognized with experience.

This paper describes the addition of nonverbal affective behaviors to a humanoid robot, as well as recognition of these behaviors based on an online survey. The expressive behaviors were implemented in the context of a framework for affective robot behavior (TAME) and span across three types of affective phenomena: traits, moods and emotions.

Indoor localization and navigation systems for individuals with visual impairments (VI) typically rely upon extensive augmentation of the physical space or expensive sensors. Thus, few systems have been adopted. This work conducts a feasibility study of whether it is possible to localize and guide people with VI using inexpensive sensors, such as compasses and pedometers, which are available in portable devices like smart phones. The proposed approach takes advantage of interaction between the system and the human user, who confirms the presence of landmarks. Experiments are employed to study what kind of directions are successful in assisting human users to reach their destination. These experiments show that Bayesian localization tools provide sufficient accuracy, while achieving real-time operation, despite the minimalistic, noisy nature of sensors and the limited computational resources available on smart phones.

We propose an unsupervised learning algorithm for the estimation of the number of components and the parameters of a mixture model. It starts from a single mixture component covering the whole data set (therefore avoiding the ill-posed problem of the components' initialization, saving also computational burden). Then, it incrementally splits that component during expectation maximization steps, thus exploiting the full space of solutions following a binary tree structure. After each component insertion it evaluates whether accepting this new solution or discarding it according with the chosen information criterion. We show that the method is faster that state-of-the-art alternatives, is insensitive to initialization (deterministic initialization strategy), and has better data fits in average. This is illustrated through a series of experiments, both with synthetic and real images.

An overlooked problem in Learning From Demonstration is the ambiguity that arises, for instance, when the robot is equipped with more sensors than necessary for a certain task. Simply trying to repeat all aspects of a demonstration is seldom what the human teacher wants, and without additional information, it is hard for the robot to know which features are relevant and which should be ignored. This means that a single demonstration maps to several different behaviours the teacher might have intended. This one-to-many (or many-to-many) mapping from a demonstration (or several demonstrations) into possible intended behaviours is the ambiguity that is the topic of this paper. Ambiguity is defined as the size of the current hypothesis space. We investigate the nature of the ambiguity for different kinds of hypothesis spaces and how it is reduced by a new concept learning algorithm.

In this paper, we propose a method for a humanoid robot to unfold clothes as a part of laundry sorting task. A clothes handling skill, which we define as a unit of robot motion, can be described as a combination of two elements: motion primitives and clothes basic information. To achieve robustness to colors and shapes of clothes, we propose clothes basic information that do not rely on color or shape of the target object. In this paper, we first propose actual motion primitives and clothes basic information for unfolding clothes, and then present experimental result obtained using a life-size humanoid robot.

This paper presents a 3D sensing system by a tracked vehicle robot using a laser range finder (LRF) with an arm-type movable unit. The unit is mounted on the robot and the LRF is installed at the end of the unit. This sensing system enables the sensor to change position and to face at a right angle corresponding to a variety of configuration. Because the occlusion can be avoided by this mechanism even for complex terrain, the robot can measure a 3D configuration such as valley, gap, upward or downward stairs more accurately than conventional 3D sensing system with the LRF, which have generally used a rotating mechanism. In this study, we have designed and developed a prototype system of the movable LRF unit and mounted it on a tracked vehicle robot with two crawlers that we have been developing. Experimental results for basic 3D sensing showed that proposed system is useful for sensing and mapping more complex terrain.

The use of ray-tracing for modeling the propagation of sound through an indoor environment has long been a mainstay of architectural acoustics. Provided accurate enough models of sound sources and the surrounding environment are available, ray-tracing can be used to predict not just volume information, but also the energy and direction of sound incident upon a receiver, and even reconstruct the wave for auralization. For mobile robots, this is important for exploring simulation work in the rapidly expanding domain of auditory perception. Given a known environment, a robot can test perception models or construct aurally dependent paths through an environment. This work evaluates the effectiveness of ray-tracing as a simulation tool through real robotic experiments, comparing the predicted sound volumes with measured data. Furthermore, this work explores using robot generated three-dimensional obstacle maps from an on-robot time-of-flight camera as part of the environment map passed to the acoustic simulator.

This paper describes a robot system which can follow a specific person while avoiding obstacles and other people walking around. This system consists of various functional modules, which are a stereo-based robust person detection and tracking, a laser range finder-based map generation, and an on-line randomized motion planning, implemented on a self-contained wheeled mobile robot. All these modules are implemented using RT-middleware, which enhances software development, maintenance, and reusability. The implemented system is tested in a reasonably complex environment with several walking people at a time. The experimental results show the feasibility of the system.

Structure-From-Motion (SFM) methods, using stereo data, are among the best performing algorithms for motion estimation from video imagery, or visual odometry. Critical to the success of SFM methods is the quality of the initial pose estimation algorithm from feature correspondences. In this work, we evaluate the performance of pose estimation algorithms commonly used in SFM visual odometry. We consider two classes of techniques to develop the initial pose estimate: Absolute Orientation (AO) methods, and Perspective-n-Point (PnP) methods. To date, there has not been a comparative study of their performance on robot visual odometry tasks. We undertake such a study to measure the accuracy, repeatability, and robustness of these techniques for vehicles moving in indoor environments and in outdoor suburban roadways. Our results show that PnP methods outperform AO methods, with P3P being the best performing algorithm. This is particularly true when stereo triangulation uncertainty is high due to a wide Field of View lens and small stereo-rig baseline.

We present a method for utilising knowledge of qualitative spatial relations between objects in order to facilitate efficient visual search for those objects. A computational model for the relation is used to sample a probability distribution that guides the selection of camera views. Specifically we examine the spatial relation “on”, in the sense of physical support, and show its usefulness in search experiments on a real robot. We also experimentally compare different search strategies and verify the efficiency of so-called indirect search.

An open challenge to the widespread deployment of mobile robots in the real-world is the ability to operate autonomously in dynamic environments. Such autonomous operation requires full utilization of the relevant sensory inputs to adapt to environmental changes. Despite being a rich source of information, vision is however, still under-utilized in robot domains because of the sensitivity to environmental changes and the computational complexity of visual input processing algorithms. This paper enables a mobile robot to better utilize the visual input to navigate safely in dynamic environments – it describes a novel algorithm that: (a) uses local image gradient cues to characterize target objects reliably and efficiently; and (b) uses temporal correspondence of visual cues for robust localization and tracking of environmental obstacles. Furthermore, the information extracted from these visual cues is merged effectively with information obtained from other visual cues and range sensors, using autonomously learned error models of the different information processing schemes. All algorithms are fully implemented and tested on a humanoid robot in dynamic indoor environments.

A cornerstone for cognitive mobile agents is to represent the vast body of knowledge about space in which they operate. In order to be robust and efficient, such representation must address requirements imposed on the integrated system as a whole, but also resulting from properties of its components. In this paper, we carefully analyze the problem and design a structure of a spatial knowledge representation for a cognitive mobile system. Our representation is layered and represents knowledge at different levels of abstraction. It deals with complex, cross-modal, spatial knowledge that is inherently uncertain and dynamic. Furthermore, it incorporates discrete symbols that facilitate communication with the user and components of a cognitive system. We present the structure of the representation and propose concrete instantiations.

Simultaneous localization and mapping (SLAM) is a fundamental problem in mobile robotics. The ever growing number of landmarks during robot action is still a problem on the way to lifelong operation. The reason is that the continuously growing number of landmarks leads to an unbounded increase of computational complexity in terms of computational power and memory.

We present a novel approach to define the benefit of a landmark in SLAM to address bounded resources. The general idea of our approach is to remove landmarks with a low localization benefit. We keep landmarks such, that their observability regions cover the operational area. The rational behind this is that the position of a landmark itself does not give a hint on its usefulness for robot localization. However, ensuring landmark visibility within the operational area leads to a minimum localization quality in the whole operational area.

We compare our approach of handling landmarks against the standard approach without upper bound on the number of landmarks in real-world experiments. These experiments are performed on a P3DX-platform with a Visual SLAM approach.

In this paper a new, efficient approach to accurate image-based homing is proposed for indoor robots that makes use of the ceiling plane. A gradient orientation histogram is built to calculate the rotation of the robot between a target location ‘snapshot’ and the current image perception, and efficient 1D average row and column images are constructed to find the vector pointing to the home location. Our algorithm is ideally suited for use on low-powered, cheap robots where sensing and computational power are limited.

Typicality problem refers to the identification of new classes in a general classification context. This typicality concept is used in this paper to help a robot acquiring a topological representation of the environment during its exploration phase. We describe a robust place recognition algorithm that fuses a number of uncertain local matches into a high-confidence global match. The problem is addressed using INCA statistic which follows a distance-based approach. Experiments performed in simulation and in a real robot/environment system show the adequateness of the approach.

In this manuscript, we pioneer an efficient indoor mobile robot navigation system using signal strength of a customized RFID system. The RFID reader is mounted on the robot and a set of RFID tags are attached to 3-D points which are known as targets to be reached by the mobile robot in an indoor workspace. First, the direction of a current target is estimated through received signal strength (RSS) measurements of the customized RFID reader. The robot's current orientation is updated to head approximately towards the target. It then applies necessary actions to its actuator to reach the current target. The path generated by the robot is better estimated using a conventional stochastic filter called Extended Kalman Filter (EKF). The customized RFID reader architecture is simulated using the comprehensive electromagnetic commercial software, FEKO. The proposed navigation system is evaluated through a number of computer simulations. It is shown through these simulations that a mobile robot has the ability to successfully navigate to reach a set of pre-defined target points in an indoor environment regardless of the 3-D positions of the RFID tags.