Mobile Robots Facing the Real World 23 in our Lab, that leads to locomotion concepts that perform extremely well in rough terrain, still being efficient and not very complex (fig and 6) 2.2 Environment Representation Environment perception and representation can be model- or behavior-based and might involve different levels of abstraction (fig 2) Whereas behavior-based approaches are often combined with bio-inspired algorithms for adaptation and Fig This figure depicts the hierarchy of abstraction levels More we go up in the hierarchy, more we reduce the geometric information and more we increase the distinctiveness For global localization and mapping, high distinctiveness is of importance, whereas for local action, precise geometric relations with the environment come forward Fig 3D representation of an indoor environment by planes The raw data (left) are registered with an upward looking SICK laser scanner, whereas a horizontally arranged laser scanner is used for probabilistic localization of the robot during the measurement Through the extraction of plans (right) from the 13157 raw data points, the representation can drastically be simplified to 27 planar regions, thus reducing memory requirement and complexity [15], and filtering unimportant information 24 R Siegwart learning, they hardly scale with more complex task and have only shown their feasibility in simple experiments Model based approaches make use of a priory knowledge by means of environment models, thus allow a very compact and taskdependant environment interpretation They scale much better with the complexity of the system and task, but have some limitations with highly nonlinear systems and unmodeled environment features However, today only model based control approaches enable predictivity of the systems behavior, thus guaranteeing safety Models can have different level of abstraction, from raw-data based grid representations up to highly symbolic descriptions of the environment (fig 2) For real world navigation, the complexity is a major issue Complexity, especially for localization and mapping, is strongly linked with the environment representation We therefore strongly belief in a model based environment representation, that drastically reduces the memory requirements and complexity Furthermore, the use of models of expected environment features enable to filter most measurement that are not relevant to the given navigation task, e.g people around the robot that are not appropriate features for localization Representing a typical office room in 2D by raw data based occupancy grids easily requires hundreds of k-bytes if a fine grid of some centimeters is used However, for localization purposes the same room can typically be represented by around 10 lines, thus requiring only around 20 to 40 bytes dependent of the requested resolution The same applies even more drastically for 3D representations as presented in figure [15] If lower precision is required, one might even use a topological map using more abstract and distinct features as the fingerprints presented in [16,17] A major challenge for a useful environment perception and representation is the inherent noise of sensor systems and the feature ambiguities that are present in most environments Therefore probabilistic algorithms are adopted in order to extract useful information and to fuse the different signals and modalities to the best estimate of the environment and the robots situation within it The currently most successful approaches employ Kalman filters [3] or Hidden Markov Models [1] for fusion and estimation The Kalman filter is well adapted if the environment is represented in a continuous form by geometric features whereas Hidden Markov Models are typically used with metric grid maps or topological maps Both approaches have their advantages and disadvantages We therefore propose hybrid approaches, using a topological representation for the global map and a metric representation based on geometric features for the local maps [18] These approaches enable to combine global consistence, robustness, precision and applicability for large environments 2.3 Navigation and Control Architecture Navigation in real world environments is a very complex task It requires an appropriate control architecture implementing various parallel tasks in real time A typical autonomous mobile robot system requires at least five control levels, running at different cycle times The main tasks ordered by importance are motor control, emergency supervision, obstacle avoidance, localization and planning of the task Apart the motor controller, all other control tasks require information about the local or even global environment of the robot As discussed in the Mobile Robots Facing the Real World 25 pervious section, the perception and representation of the environment can become very complex Thus the processing power to run these algorithms can be extremely high and therefore real-time implementation a real challenge Various research projects address this problem with the goal to find new concepts and algorithms for robust and practical navigation in real world environments [1,3] Today, feasible solutions for typical indoor or flat outdoor environments are available (e.g RoboX presented below) However, navigation in unstructured and rough terrain, where the environment has to be modeled in real 3D, is still a very open research area [4,5,6] Examples At the Autonomous Systems Lab at EPFL we conduct focused research in mobile robotics for autonomous operation in real world environments Major axes are in the field of mobile robot design for rough terrain, navigation and interaction, and in mobile micro-robotics Among our most recent findings are enhanced feature based localization concepts [3,9,8], obstacle avoidance for highly dynamic and human-cluttered environments, wheeled robots for high performance in rough terrain [12,4,10] and a mobile micro-robot of the size of a sugar cube [19] Recently we run one of the worlds largest mobile robot installations with eleven fully autonomous and interactive mobile robots at the Swiss exhibition expo.02 It represents a milestone in our mobile robotics research and allowed us a long-term evaluation of our recent findings Furthermore it was used to investigate social and psychological issues of mobile robotics In the following two research results are briefly presented and discussed 3.1 Wheel-Based Mobile Robot Locomotion for Rough Terrain Wheels enable efficient motion on flat ground, and, if equipped with an appropriate suspension, can reach excellent climbing abilities In our Lab we therefore investigate new passive and active wheel-based locomotion concepts Passive means, that the articulations of the suspension have no actuators, whereas active concepts use motors for at least some of the articulations of the suspension system Shrimp, presented in figure 4, is a passive system with wheels [12] It can effortlessly overcome obstacles up to two times of its wheel diameter and climbs regular steps of stairs that are about of its height Within a running research project for the European Space Agency, the system is extended for full energetic autonomy using solar cells, and equipped with a navigation system for autonomous long-range operation [4] Octopus, shown in figure 5, features an active locomotion concept on wheels Its active and one passive articulations enables the robot to keep all wheels in optimal ground contact at any time [10] A specially developed tactile wheel measures the contact point and force of each wheel 26 R Siegwart Fig The robot Shrimp is an all-terrain rover based on a passive locomotion concept It is characterized by wheels suspended by parallel mechanisms, one fixed wheel in the rear, two boogies on each side and one front wheel with spring suspension The robot sizes around 60 cm in length and 20 cm in height, is highly stable in rough terrain and overcomes obstacles up to times its wheel diameter with a minimal friction coefficient Fig Octopus features an active locomotion concept with motorized and tactile wheels, active and passive DOF for ground adaptation and on-board integration of all control elements, joint sensors and inclinometers (Photo © Bramaz) Mobile Robots Facing the Real World 27 3.2 RoboX, the Tour-Guide Robot with Long-Term Experience [11] The Swiss National Exhibition takes place once in 40 years The 2002 edition, expo.02, ran from May 15 to October 21, 2002 It hosted the exhibition Robotics that was intended to show the increasing closeness between man and robot technology (fig 1) The central visitor experience of Robotics was the interaction with eleven autonomous, freely navigating mobile robots on a surface of about 315 m2 Their main task was giving guided tours but included also a robot taking pictures of visitors The exhibition was scheduled for five hundred persons per hour For this task, the main specifications can be summarized as follows: • Navigation in an unmodified, highly populated environment with visitors and other freely navigating robots • Bi-directional multi-modal interaction using easy-to-use, intuitive yet robottypical interaction modalities • Speech output in four languages: French, German, Italian and English • Safety for visitors and robots at all time • Reliable operation during around eleven hours per day, seven days per week, during five months • Minimal manual intervention and supervision • Adaptive multi-robot coordination scenarios in function of the number of visitors and their interests • Control of visitor flow through the robots • Development of ten robots within tight budgets and schedules The RoboX robot was designed and developed at our Lab by a multidisciplinary team of around 15 young engineers and artists (fig and 7) It was then realized by our spin-off company BlueBotics It features fully autonomous navigation, including feature-based localization [3], highly adaptive and dynamic obstacle avoidance [8] and multi robot coordination on the path planning level [2] The main interaction functions are face and people tracking, speech output, facial expression through the two pan-tilt eyes and the eye-integrated LED matrix [11] Four touch buttons were used as input devices and two robots were equipped with a directional microphone and speech analysis for simple answers The navigation and interaction software, with around 20 main tasks, was running on two embedded computers The safety-critical navigation software runs on a XO/2 operating system based on Oberon [13] and the interaction software on an industrial PC running Windows 2000 An additional security controller was running on a PIC micro-controller, guaranteeing visitors safety at all time The specially developed interaction software SOUL [20] aimed at composing the scenarios like a theater or a music composition It enables through a convenient interface to combine different basic behaviors with synthesized speech, motion, sensory inputs and much more 28 R Siegwart Fig a) The autonomous exhibition robot RoboX b) RoboX number with visitors in the pavilion at expo.02 Fig Basic elements and functionalities of the tour-guide robot RoboX Eleven robots were guiding visitors through the exhibition and interacting with them in an environment cluttered by hundreds of visitors During the fivemonth exhibition, the RoboX family was in contact with 686'000 visitor and traveled a total distance of 3'315 km The installation served also as research Mobile Robots Facing the Real World 29 platform and technology demonstration Throughout the five-month operation period, the navigation system was close to 100% reliable This was especially due to the localization system that was based on line feature [2], thus filtering out all the dynamics in the environment coming from the visitors in the vicinity of the robot More details on the hardware design, the navigation system and the reliability can be found in [2,8,11,14] Conclusions and Outlook New concepts for wheeled locomotion, feature based environment representation and navigation have been presented and discussed in this paper Their potential was shown by two examples of mobile robots system of our Lab, facing the complexity of the real world They represent our recent findings, but are still only a very first step towards intelligent and socially interactive robots In order to realize really intelligent mobile robots, able to scope with highly complex real world environments, enormous research efforts in various fields like environment representation, cognition and learning are still required Acknowledgments The author would like to thank all the current and past collaborators of the Autonomous Systems Lab for their contributions and inspiring work, their curiosity and dedication for mobile robotics research The presented projects were mainly funded by EPFL, the European Space Agency (ESA) and expo.02 References S Thrun, D Fox, W Burgard, and F Dellaert, "Robust Monte Carlo Localization for Mobile Robots," In Artificial Intelligence (AI), 2001 Arras, K.O., Philippsen, R., Tomatis, N., de Battista, M., Schilt, M and Siegwart, R., "A Navigation Framework for Multiple Mobile Robots and its Application at the Expo.02 Exhibition," in Proceedings of the IEEE International Conference on Robotics and Automation (ICRA'03), Taipei, Taiwan, 2003 Arras, K.O., Castellanos, J.A and Siegwart, R., Feature-Based Multi-Hypothesis Localization and Tracking for Mobile Robots Using Geometric Constraints In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA’02), Washington DC, USA, May 11 - 15 2002 Lamon, P and Siegwart, R., "3D-Odometry for rough terrain – Towards real 3D navigation." In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA'03), Taipei, Taiwan, 2003 Singh S., Simmons R., Smith T., Stentz A., Verma V., Yahja A., Schwehr K., "Recent Progress in Local and Global Traversability for Planetary Rovers", Proceedings of IEEE International Conference on Robotics and Automation (ICRA’00), p1194-1200, San Francisco, April 2000 A Mallet, S Lacroix, and L Gallo, "Position estimation in outdoor environments using pixel tracking and stereovision", Proceedings of IEEE International Conference on Robotics and Automation (ICRA’00), pages 3519-3524, San Francisco, CA (USA), April 2000 Fong T., Nourbakhsh I., Dautenhahn K., "A survey of social interactive robots," Journal of Robotics and Autonomous Systems, 42, 143-166, 2003 30 10 11 12 13 14 15 16 17 18 19 20 R Siegwart Philippsen, R and Siegwart, R., "Smooth and Efficient Obstacle Avoidance for a Tour Guide Robot," In Proceedings of IEEE International Conference on Robotics and Automation, (ICRA’03), Taipei, Taiwan, 2003 Tomatis, N., Nourbakhsh, I and Siegwart, R., "Hybrid Simultaneous Localization and Map Building: Closing the Loop with Multi-Hypotheses Tracking," In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA’02), Washington DC, USA, May 11 - 15, 2002 Lauria, M., Piguet, Y and Siegwart, R., "Octopus - An Autonomous Wheeled Climbing Robot," In Proceedings of the Fifth International Conference on Climbing and Walking Robots Published by Professional Engineering Publishing Limited, Bury St Edmunds and London, UK, 2002 Siegwart R., et al., "Robox at Expo.02: A Large Scale Installation of Personal Robots," Special issue on Socially Interactive Robots, Robotics and Autonomous Systems 42 (3-4), 31 March 2003 Siegwart R., Lamon P., Estier T., Lauria M., Piguet R., "Innovative Design for Wheeled Locomotion in Rough Terrain," Journal of Robotics and Autonomous Systems, Elsevier Sep 2002, Vol 40/2-3, pp 151-162 Brega, R., N Tomatis, K Arras, and Siegwart R., "The Need for Autonomy and RealTime in Mobile Robotics: A Case Study of XO/2 and Pygmalion," IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS’00), Takamatsu, Japan, 2000 Siegwart, R., Arras, K.O., Jensen, B., Philippsen, R and Tomatis, N., "Design, Implementation and Exploitation of a New Fully Autonomous Tour Guide Robot," In Proceedings of the 1st International Workshop on Advances in Service Robotics (ASER'2003), Bardolino, Italy, 13-15 March 2003 Weingarten, J., Gruener, G and Siegwart, R, "A Fast and Robust 3D Feature Extraction Algorithm for Structured Environment Reconstruction," Proceedings of 11th International Conference on Advanced Robotics, Portugal, July 2003 Lamon, P., I Nourbakhsh, et al., "Deriving and Matching Image Fingerprint Sequences for Mobile Robot Localization," Proc of IEEE International Conference on Robotics and Automation (ICRA), Seoul, Korea, 2001 Lamon, P., Tapus A., et al., "Environmental Modeling with Fingerprint Sequences for Topological Global Localization" - submitted at IROS’03, Las Vegas, USA, 2003 Tomatis N., Nourbakhsh I and Siegwart R., "Hybrid Simultaneous Localization and Map Building: Closing the Loop with Multi-Hypotheses Tracking." In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA’02), Washington DC, USA, May 11 - 15, 2002 Caprari G., Estier T., Siegwart R.: "Fascination of Down Scaling - Alice the Sugar Cube Robot, Journal of Micro-Mechatronics," VSP, Utrecht 2002, Vol 1, No 3, pp 177-189 Jensen, B., Froidevaux, G., Greppin, X., Lorotte, A., Mayor, L., Meisser, M., Ramel, G and Siegwart, R (2003) "Multi-Robot Human-Interation and Visitor Flow Management," In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA’03), Taipei, Taiwan, 2003 Breakthroughs in Human Technology Interaction Bernd Reuse Federal Ministry of Education and Research, Germany Abstract In 1999 the German Federal Government launched six major strategic collaborative research projects on Human Technology Interaction, which involved 102 research partners and a funding volume of 82 million The results of these projects were expected to allow people to control technical systems multimodally by using natural forms of interaction such as speech, gestures, facial expressions, touch and visualization methods and to apply such systems for the most varied purposes in their private and working environments The ambitious research goals were achieved with prototypes for real-world applications Research activities have resulted in 116 patent applications, 56 spin-off products and 13 spinoff companies as well as 860 scientific publications Introduction: Trends in Human Technology Interaction Together with the Federal Ministry of Economics and Labour (BMWA), the Federal Ministry of Education and Research (BMBF) organized an international status conference in Berlin in June 2003, where the results of four years of governmentfunded research on Human Technology Interaction (HTI) were presented (www.dlr.de/pt-dlr/sw) Distinguished personalities from science, research and industry participated in this conference The roughly 350 conference participants from Germany and abroad agreed on the following trends in Human Technology Interaction: • Human Computer Interaction is turning into Human Computer Cooperation and will support many trends in I&C The number of transactions is increasing dramatically and requires new modalities in HTI Access to any information with any device at any place and at any time will be supported by HTI interfaces Agents will take over routine work • A simple and easy-to-handle Human Computer Interface is an important precondition for marketing the products of the IT industry • Human Computer Interaction can help solve the problems of the future, namely the problems of the aging society An intelligent human life has to be supported by science and technology S Yuta et al (Eds.): Field and Service Robotics, STAR 24, pp 31–38, 2006 © Springer-Verlag Berlin Heidelberg 2006 32 B Reuse • Human Computer Interaction will have a strong influence on society as a whole as a result of the convergence of the use of computers in private and business environments, of working life and leisure time, and of paid and unpaid work The Verbmobil Project After 20 years of largely unsuccessful research in the field of speech recognition, which had only produced simple dialogue systems, the BMBF decided in 1993 to provide a total of approximately 60 million for an eight-year research project entitled Verbmobil which was to deal with the automatic recognition and translation of spontaneous speech in dialogue situations This was not in line with the then prevailing trend in research, and even some of the international experts who were involved took the view that the goal could not be achieved at that time But in the Verbmobil project we pursued new paths in research: first of all, mastering the complexity of spontaneous speech with all its phenomena such as vagueness, ambiguities, self-corrections, hesitations and disfluencies took priority over the envisaged vocabulary Another novel feature was that researchers used the information contained in the prosody of speech for speech recognition purposes In addition, the transfer included knowledge processing, which is indispensable in translation The 135 different work packages and 35 research groups distributed throughout Germany were linked by a network management led by Professor Wahlster of the German Research Centre for Artificial Intelligence (DFKI) in Saarbrücken As a result of Verbmobil it was possible to demonstrate in July 2000 the translation of spontaneous speech for the domain of the remote maintenance of PCs using 30,000 words and for a telephone translation system with 10,000 words for translation from German into English and with 3,000 words for translation from German into Japanese In addition Verbmobil generated 20 spin-off products, spin-off companies and about 800 scientific publications In 2001, Verbmobil received Federal President Rau's German Future Award, the highest German research award Lead Projects on Human Technology Interaction In 1999 the Federal Government started an initiative on Human Technology Interaction, the central goal being to extend the findings and models of the Verbmobil project concerning speech-based human interaction with computers to cover the full range of human forms of interaction It was expected that the consideration and integration of several forms of interaction would allow a much better interpretation of the user's intention than one Indoor Navigation for Mobile Robot 45 The robot is given a command of goal information through wireless LAN from an operator On basis of start and goal information, the robot plans a rough path with referring a topological map by using graph search method The robot starts to move The robot runs along wall in a passage direction of the goal according to a path planning The robot can also detect a obstacle by infrared sensor, and avoid it When the robot arrives at a junction, it will detect the response from IA Then, the robot communicates with IA and requests the guidance in the junction Figure shows a communication process between the robot and the IA The IA receives a start and goal information from the robot, and give a trajectory information to the robot with referring a local information The IA gives a laser spot as a target point for the robot movement by using OP on basis of the trajectory The robot detects a laser spot from OP which controlled by IA, then moves to the spot IA guides a robot with moving a laser spot along the trajectory in a junction 10 When the robot arrives a another passage, the robot sends a arrival message to the IA Then, the guidance by IA is finished 11 The robot starts an along wall run again 12 When the robot puts in another junction, the optical guidance is executed If the robot detect a goal information form IA, the task is finished In addition, in the case of many robot move into the junction simultaneously, the robot which communicated first with IA receives guidance service Other robots receives a busy signal from IA, then wait on current position until service becomes possible Therefore, navigation is possible even if two or more robots exist in the environment Navigation System In order to realize a proposed navigation, the prototype system is developed This section describes a details of the system 4.1 Omni-Directional Mobile Robot The omni-directional mobile robot ZEN has been already developed for realizing flexible action[5](Fig (a)) The control system is mounted on the robot Batteries are also mounted on the robot for electrical devices and actuators The robot can behave autonomously and independently The robot has an infrared sensor system called LOCISS (Locally Communicable Infrared Sensory System)[6] shown in Fig.5 (b) The robot can detect a wall and run along wall by using LOCISS, and can also detect a obstacle and avoid it 46 T Suzuki et al Fig Communication between robot and IA Fig Omni-directional mobile robot 4.2 Information Assistant and Optical Pointer Figure (a) shows a Intelligent Data Carrier (IDC)[7] and IDC Reader/Writer IDC is a small device which consists of radio communications part, CPU, memory and battery The robot can communicate with IDC through IDC Reader/Writer In this research, we used this IDC ver.4 as IA, and installed in the environment For local navigation in a junction, OP (Optical Pointer) was developed Figure (b) shows a OP which is installed in the ceiling OP consists of two stepping motors for pan- Indoor Navigation for Mobile Robot 47 tilte motion, two motor driver for driving stepping motor and a laser pointer OP is connected to IA by the cable IA controls the actuators and laser pointer of OP It is enabled to give a laser spot flexibly on the floor surface Fig IA and OP Experiments The proposed system was installed to an indoor environment where was the second floor of our building in RIKEN Figure (a) shows the floor map of experimental environment The task of the robot is movement from its present location to a goal which is given by the human IA and OP are installed in all junction of the floor They have information about own position and junction state The circles A, B and C of Fig (a) show junctions where the IA-OP system is installed and also show communication and local navigation range of each IA-OP system Experimental result is also shown in Fig The robot had information of current position as START, and was given a goal such as shown in Fig (a) Then, the robot planed a rough path such like START - A - B - C - GOAL with referring a topological map by using graph search method After that, the robot run along the wall based on a rough path by using LOCISS, and can also avoid a obstacle When the robot approached to a junction, the robot detected and communicated with IA The IA projected a laser spot onto a floor surface as a target point for the robot movement by using OP The IA provided a trajectory to the robot by moving a laser spot Figure (b) shows a trajectory which is given a laser spot movement in junction A Likewise, the robot could reach the goal by using both of an along wall run based on rough path planning by using topological map and the local navigation in junction B and C The dotted line in Figure (a) shows the actual trajectory of the robot Figure shows a situation of local navigation in junction 48 T Suzuki et al Fig Experimental Result Fig Robot navigation experiment using IA and OP Conclusion The new navigation strategy for mobile robots operating in indoor environment is proposed The robots need a static and global information describing a topological map such as positional relation from any starting position to any goal position for making a rough path plan as well as dynamic and local information including local map, obstacles, traffic information for accurate navigation control For intelligent navigation in indoor environment, we classified the information into two types such as global information and local information and proposed the method for managing each type using the IA system and the OP The navigation algorithm by using the system was presented The experimental example of navigation by using this system is shown The robot could navigate to the goal efficiently by using the systems ... static and global information S Yuta et al (Eds.): Field and Service Robotics, STAR 24 , pp 41–49, 20 06 © Springer-Verlag Berlin Heidelberg 20 06 42 T Suzuki et al describing a topological map [2] such... 20 02, Vol 40/ 2- 3 , pp 15 1-1 62 Brega, R., N Tomatis, K Arras, and Siegwart R., "The Need for Autonomy and RealTime in Mobile Robotics: A Case Study of XO /2 and Pygmalion," IEEE/RSJ International... society An intelligent human life has to be supported by science and technology S Yuta et al (Eds.): Field and Service Robotics, STAR 24 , pp 31–38, 20 06 © Springer-Verlag Berlin Heidelberg 20 06