Mobile robots are becoming more and more sophisticated and the research on them is turning towards new applications. One of the desired advancement of mobile robots nowadays is their deployment in spaces that are populated by humans, like homes, offices, public spaces, etc. This will make possible new applications, such as robot servants or other service robots. However up to this date, there are almost no working implementations of robots in such environments. The reason for that lies in the fact that the robotic intelligence has still not matured to a level that would allow easy coexistence with humans. Computers are not yet able to grasp and deal with the complex organization of our lives, and the resulting complexity of our actions and the environments we live in.

The standard approach to the application of mobile robots is to provide the robot with sensors and computing power, and implement the necessary intelligence to do the required tasks, much in the same way like humans do. However, seeing that using this approach we are still not able to achieve the needed smartness and abilities of the mobile robot, in this thesis we propose a different approach to the problem. This is to distribute the intelligence between the robot and the space it acts in, under the assumption that this kind of synergic combination of outside and inside intelligence will lead to a considerable advantage when compared to the standard all-onboard approach.

Therefore, in the thesis we work on the control of the mobile robot in environments that have distributed sensing and computing power. These types of spaces are given the name Intelligent Spaces. Although the research on such spaces is steadily growing, its application for better control of mobile robot has not yet been discussed. This is the objective of this thesis. The thesis is divided into several chapters, which present parts of the work done during the doctoral course.

Chapter 2 gives an overview of what Intelligent Spaces are, what are their characteristics and what can be done with them. As a fairly new research ground there it has not progressed very much, however the benefits of having distributed intelligence in the space are steadily catching the interests of many researches. The number of applications on the human understanding or human machine interaction is steadily rising, and it is expected that this kind of concept will become standard in that area. An overview of similar researches is also given, describing the basic trends and similar research areas, such as ubiquitous computing or sensor networks. After that the connection between mobile robots and Intelligent Spaces is described. Here we discuss what is that mobile robots offer when implemented in Intelligent Spaces - this can be broadly be divided into two applications: media for physical services and mobile sensing device. Following that, we discuss the main merits the introduction of Intelligent Space has for mobile robots, however a detailed discussion on that topic is reserved for later chapters.

The following chapter 3 is concerned with the motion control of the mobile robot. Considering the type of application we are aiming at in this work, here we consider several methods for robot control and compare their characteristics. The control is divided into two parts: global path planning and local path following and obstacle avoidance. Each of these parts is explained and the methods explained and compared in experiments. Then path planning and obstacle avoidance - i.e. the methods which gave the best result: Field D* method and the Dynamic Window approach - are combined together in order to get a control method that has good navigation characteristics and can achieve movement at high speeds with consideration of obstacles or humans around the robot and without bumping into them. The basic two-step method is also expanded in order to achieve better behavior in the vicinity of dynamic objects. Since in the environments we are trying to make are robots move there will be many dynamic objects, mostly humans, this type of improvement is needed. We describe a rather simple but effective moving object avoidance algorithm based on the prediction of the future position.

The methods described in chapter 3 rely heavily on the measurement and estimation of the relevant variables, which in this case are the position and speeds of both the mobile robot and the humans in the space, and the position of walls and static objects - in other words the map of the space. These tracking and mapping tasks are dealt with in chapter 4, where sensing inside Intelligent Spaces is described. Here the accent is on the deployment of sensors that are distributed and fixed at different locations in the space. A description of mainly used sensors for this type of applications is given, and afterwards the details of the implementation of tracking using the ultrasound positioning system and multiple laser range finders are described.

The ultrasound system can only be used as is, and due to its relatively complex installation and other factors it is probably not a good solution for real spaces, except perhaps for specific situations. However it gives a testbed for other sensing system due to its good accuracy. Laser range finders on the other hand are easy to be applied in everyday environments, due to their small size, accuracy and easy installation. But, a laser range finder as it is does not provide tracking or mapping abilities. We describe the developed method for tracking and mapping, which allows the sensors to be quickly installed and connected together to give a complete sensing system. This way any space can be easily turned into an Intelligent Space. Apart from the basic method for tracking we also discuss mapping, calibration and placement of the sensors. Another important part of the tracking system is the ability to distinguish between types of objects (i.e. humans or robots), and the implementation of that function is also described in this chapter.

In chapter 5 sensing using the mobile robot's onboard sensors in addition to the distributed sensors of the Intelligent Space is discussed. The inclusion of a mobile sensor into the network of static sensors from chapter 4 is not straightforward, as the two types of sensors have different characteristics. This is analyzed here, and the main factor is pointed out: the motion of the mobile robot introduces correlations in the estimate (similar to the SLAM problem), which in turn can result in an increase of the computational and communicational burden of the tracking system. In order to avoid this, we introduce the use of the Covariance Intersection method, in which case it is possible to combine the two types of sensors directly, however at the cost of a higher uncertainty in the estimate, than in the case when not using this method. Also here we analyze several different types of information fusion architectures that can be used for such a combined sensor tracking system. Analyzed are centralized, decentralized and distributed architectures, and the modalities of their application are considered. The main reason for choosing one or the other lies in the fact that mobile robots are present in the system.

Finaly, a different method for tracking with both distributed and onboard sensors that uses a geometric model of both the robot and the environment is presented. The tracking method is based on particle filters, and implementations of both simple tracking and tracking and model building (or mapping) are presented. A comparison with the first method is given.

In chapter 6 the results from the previous chapters are combined in order to obtain a complete mobile robot control system based on the cooperative sensing using both onboard sensors and sensors distributed in the space. Here we discuss the proposed control scheme and analyze what are the advantages that can be realized when external sensors from Intelligent Space are added in order to enhance the overall sensing abilities needed by the mobile robot. The parts of the control system that are improved the most are the ability to understand the humans' movement and predict their future plans through constant observation, the ability to "see" and measure what would usually be impossible only with onboard sensors and also a large improvement in the robot localization and object tracking characteristics. We also argue that this type of implementation greatly improves the robot's characteristics, to the point that its introduction into populated and dynamic spaces of our everyday life becomes feasible and even easy.

The discussion is followed by experimental results, where the merits of the proposed scheme are demonstrated in several different situations. The experiments show that the novel scheme indeed has a great influence on the robot's performance.

The thesis concludes with a list of contributions, plans for future work and a comment on the perspective of mobile robots that move and work in our vicinity.

本論文は、「分散されたセンサとオンボードセンサに基づく移動ロボットの制御スキームに関する研究」と題し、全6章から構成され、人の日常環境に移動ロボットを導入するために環境に分散配置されたセンサとロボットのオンボードセンサの情報を統合する手法及びロボット制御のために必要な制御スキームを提案し、それらを実験によって実証し有効性を明らかにしたものである。

第1章では、「序論」と題し、移動ロボット研究の現状を踏まえ人間と共存した環境における移動ロボット制御の問題点を指摘し、それを解決するため空間知能化技術との統合を本研究の目的として述べている。

第2章では、「空間知能化と移動ロボット」と題し、空間知能化とその関連研究を紹介し、特に空間知能化における移動ロボットの役割とそれを実現するための制御スキームについて述べている。更に、先行研究におけるセンシング機能と研究課題の考察を通じて移動ロボット制御に必要なセンシング課題について明らかにしている。

第3章では、「移動ロボット制御」と題し、経路計画、経路追従・障害物回避といった機能から構成される移動ロボットの制御手法に関して述べている。様々な手法の比較検討を行ない、本論文では経路計画手法としてField D*、経路追従・障害物回避手法としてDynamic Windowを採用し、実証実験を通してその有効性を示している。更に、動的な環境に対応するために移動物体(人間)の行動を予測し経路計画に反映させる手法を提案している。

第4章では、「分散されたセンサによるセンシング」と題し、空間知能化における移動物体トラッキングに用いられる様々なセンサを紹介し、それぞれの特徴について述べている。特に超音波による位置計測装置を用いたトラッキング手法について、その問題点を明らかにしている。その考察をもとに本研究ではトラッキングのためのセンサとしてレーザレンジファインダを適用することにした。レーザレンジファインダを用いた人とロボットの位置トラッキング手法を提案し実証実験を通して有効性を示している。一方、同様に移動ロボット制御に必要な情報として環境地図があげられる。しかしその構築に関しては分散されたセンサを直接用いるだけでは不十分であることを明らかにしている。

第5章では、「分散されたセンサとオンボードセンサによるセンシング」と題し、第4章で明らかになった問題点を解決するために分散されたセンサに加え移動ロボット上のセンサを用いることを提案している。初めに両者の情報を統合することで人間とロボットのトラッキング及び環境地図の構築を精度良く行えることを、実験を通して示している。この統合においては測定値の間に相関関係が生じるため計算量・通信量の増大に対処する必要がある。そこでCovariance Intersection手法による統合を行い、相関関係の問題が解決できることを示している。更に、相関関係の影響を削減するいくつかのヒューリスティックな手法を提案している。最後にパーティクルフィルタを用いて移動ロボットモデルを獲得しそれに基づく統合手法を提案し、よりロバストにかつ広い適用範囲で使用できることを示している。

第6章では、「結論」として、本研究で得られた成果をまとめ、残された問題と今後の研究方向を述べている。

以上を要するに、本論文は、人の日常環境に移動ロボットを導入するために、環境に分散配置されたセンサとロボットのオンボードセンサの情報を統合する手法、及びロボット制御のために必要な制御スキームを提案し、それらを実験によって実証し有効性を明らかにしたものであり、電気工学、ロボット工学に貢献することが少なくない。よって、本論文は、博士(工学)の学位請求論文として合格と認められる。