THE history of Unmanned Ground Vehicles (UGVs)

Transcription

1 The Evolution of Vision-based Unmanned Ground Vehicles: the VisLab Perspective Massimo Bertozzi, Alberto Broggi, and Alessandra Fascioli VisLab - Dip. di Ingegneria dell Informazione Università degli Studi di Parma Parco Area delle Scienze, 181A, Parma, Italy {bertozzi,broggi,fascal}@ce.unipr.it 1 Unmanned ground vehicles are one of the enlightening ideas that will shape our future by providing enhanced safety and improved mobility. Although the development of prototypes is still going on, we are now assisting to demonstrations that witness the real applicability of the early concepts. THE history of Unmanned Ground Vehicles (UGVs) developed in the last years. In fact, although the first ideas were born in the sixties, the level of maturity of the technology at that time did not allow to afford the original goal of implementing fully autonomous all-terrain all-weather vehicles. The first documented prototypes of automated vehicles were developed by a few groups in the military arena in the mid eighties. The initial stimulus that had triggered these innovative ideas had been provided by the military sector, who was eager to provide complete automation to its fleet of ground vehicles. It was not before the eighties that this interest was then transferred to the civil sector: governments worldwide launched the first projects which conveyed a large number of researchers over these topics. But the interest of the automotive industry in developing real products - and thus investing money in this new business- started to develop only when the feasibility studies were successfully over and the first prototypes were demonstrated. The test of autonomous vehicles on real roads in a real environment was one of the most important milestones in the history of unmanned ground vehicles. This happened in the mid-late nineties. We are now assisting to a flourishing of research initiatives oriented to intelligent vehicles. Anyway, due to many technological issues and legal considerations, full autonomy was not set as a goal: the automotive industry feels as its primary goal the need to equip vehicles with supervised systems and -more generally- ADAS (Advanced Driving Assistance Systems) instead of automatic pilots. At the same time the research on UGVs then slowed down, since it was not felt as a primary strategic area of investment any more by neither industry nor governments. Departments of Transportation worldwide were in fact primarily interested in social, economic, or environmental objectives aimed at enhancing the efficiency of fuel, road network, and improving the quality of life in terms of mobility. The good results reached by ADAS in the car market in the last years induced the military sector to give a new vigorous push to the original ideas of automating its fleet of ground vehicles. The Defense Advanced Research Projects Agency (DARPA) Grand Challenge was the winning idea that attracted a large number of top-level research institutes who worked with the million-dollars prize in mind and helped/brought the scientific society to a considerable step forward. Indeed the description of the applications of unmanned ground vehicles given above is far from being complete, since unmanned vehicles technology is being considered for many other applications. The automation of road vehicles is perhaps the most common one and the one that attracts the largest interest from the industry. However, other domains such as agricultural, demining, rescue, and dangerous applications in general, are looking at autonomous vehicles as a possible solution to the issue of ever-increasing personnel cost. If a vehicle could move autonomously on a field to seed, or enter in a mined field, or even perform dangerous missions, the number of individuals put at risk would drastically decrease and

2 at the same time the efficiency of the vehicle itself would be increased thanks to a 24/7 operational schedule. However, this paper deals with the most common applications of intelligent vehicles, and -as mentioned above- the domain which is currently catalyzing the interest of the automotive industry, car makers, and providers of automotive appliances. The paper is structured to describe two parallel stories which intersect and share some important milestones. On one side the history and evolution of intelligent vehicles worldwide and, on the other side, the history of the Artificial Vision and Intelligent Systems lab at the University of Parma (VisLab). The evolution of VisLab in terms of projects, funds sources, and objectives is -in fact- very similar to and follows very closely the developments obtained worldwide in the same field [2]. For this reason, the description of the evolution of intelligent vehicles is here supported by real examples and case studies conducted at VisLab throughout the last 15 years. VisLab is a very focussed laboratory which has been doing research in the field of artificial vision for intelligent vehicles for 15 years and has obtained a relevant position in the global scenario thanks to the milestones that VisLab itself set in the history of intelligent vehicles. The issues that are covered by this review are primarily considerations on (i) the ideas, (ii) the enabling technologies or limitations that in one way or another conditioned the research, (iii) the main challenges that triggered the interest towards this interdisciplinary research stream worldwide, and (iv) the lessons learned by both the scientific community and the researchers at VisLab. I. PRE-COMPETITIVE RESEARCH Following the first experiments sponsored by military organizations and involving both ground and aerial applications, at the end of the eighties and the beginning of the nineties, the interest of the scientific community on Intelligent Transportation Systems (ITS) was beginning, involving academic and private research centers in a precompetitive research stage. Several national and international organizations launched and funded research programs in the ITS field to reduce traffic congestion and to increase safety. The Mobility 2000 program and the Automated Highway System project in the United States, the RACS and AMTICS projects in Japan, the DRIVE and EUREKA programs in Europe, are examples of such initiatives. Due to the real-time requirements of systems to be installed on board of a vehicle, the development of custom architectures was a mandatory approach for most research groups. These architectures were obtained both assembling off-the-shelf components or, rather, starting from the silicon. The use of devices like radar or acoustic sensors able to directly measure quantities of interest was a common choice, since the small amount of data fed by these devices did not require high computational capabilities. Conversely, vision sensors acquire a large amount of data and require a complex processing phase. Nevertheless, the use of vision, even if challenging, was indeed considered, since it can provide a rich description of the environment without requiring specific road infrastructures. The following paragraph describes the early projects carried out by VisLab at the beginning of the nineties. The MOB-LAB vehicle: The MOB-LAB (MOBile LABoratory) vehicle was developed within the framework of the PROMETHEUS (PROgraM for a European Traffic with Highest Efficiency and Unprecedented Safety) project, a pan-european EUREKA program ( ) initiated by the automotive and automotivesuppliers industry in Europe. The MOB-LAB prototype (a FIAT Ducato 18 Maxi) was a laboratory suited to study, develop, and test real time intelligent systems also using computer vision systems. It was equipped with B/W and Color analog cameras in the front and the back of the vehicle to analyze the external environment. A fully custom architecture based on the SIMD (Single Instruction Multiple Data) paradigm was developed. This system (named PAPRICA, PArallel PRocessor for Image Checking and Analysis) was directly interfaced to a stereo vision system and able to directly display the results of the processing on an external monitor. Moreover, PAPRICA featured a hardware support for pyramidal-based image processing and was therefore suitable for multi-resolution image processing. The early solutions and algorithms conceived for the two basic functionalities (lane and vehicle detection) were tested on MOB-LAB. Lane detection was based on the use of a lane model at different resolutions to exploit the pyramidal capabilities of the PAPRICA architecture. Similarly, a rectangular shaped template was used for obstacle detection. These initial results, while being at an embryonal stage, demonstrated that the use of vision for intelligent vehicles was a viable path. Thanks to the use of a 2

3 dedicated hardware the two detection systems were able to run in a real-time fashion. In both cases, grey-level images were processed; the use of color images was tested but it required the processing of a larger amount of data and was therefore unaffordable. Initially, algorithms were tested on the massively parallel Connection Machine CM-2 and then ported onto the PAPRICA architecture. In addition, the use of a dedicated architecture required also to develop a operating environment consisting of code assembler, debugger, and graphic libraries. The PAPRICA system demonstrated that the inadequacies of computers available at that time could be solved developing custom architectures. The SIMD paradigm implemented on PAPRICA was a highly successful choice witnessed by the following Intel adoption of a similar solution: the MMX extensions, a SIMD extension for images and sound processing. II. PIONEER PROJECTS Given the promising perspectives opened by the precompetitive phase, a few research centers gave birth to pioneer projects aimed at the development of concept vehicles to be later transformed into real prototype of intelligent vehicles. Although the scientific community were increasing their awareness of the potentialities of ITS, in this explorative phase the interest of the automotive industries was not born yet. The research efforts mainly derived from the individual initiative of single groups or from institutions willing to invest in research areas characterized by a high degree of risk. Two outstanding examples of such pioneer projects were the NavLab vehicle developed at the Carnegie Mellon University, U.S.A. [6], and the VaMoRs prototype demonstrated by the Universität der Bundeswehr, Germany [5]. A common choice in these projects was the use of machine vision as the main sensing device. The technological solutions available for the processing system were different: some research groups chose a general purpose hardware, some turned to a completely customized processing engine, while others adopted a mixed solution. The main challenge faced in these projects was the implementation and on-field testing of prototype vehicles equipped with vision-based driving assistance systems. The following sections describes two examples of pioneer projects carried on by VisLab in the late nineties, the ARGO and RAS projects. The ARGO project: ARGO [3] was one of the first prototype vehicles showing autonomous driving capabilities tested on public roads. It integrated the main results of the research conducted over the preceding few years on the algorithms and the architectures for vision based automatic road vehicles guidance. A number of different solutions for autonomous navigation had been developed, tested and tuned, particularly for the basic functionalities of Obstacle Detection and Lane Detection. The most promising approaches for both functionalities were integrated into the GOLD (Generic Obstacle and Lane Detection) system [1] which acted as the automatic driver of ARGO. Obstacles on the road were detected and localized by means of stereo vision, while the processing of a single monocular image allowed to extract the road geometry in front of the vehicle. The two functionalities shared the same underlying approach, the Inverse Perspective Mapping, to remove the perspective effect from the acquired image. Lane detection was performed through the extraction of specific image features, followed by the use of models of the lane markings. The application of IPM to stereo images, in conjunction with an a-priori knowledge on the road shape, also allowed to compute the free space and detect generic obstacles through a simple match with a model representing the environment without obstacles. ARGO was a passenger car equipped with a vision system composed of two low-cost B/W cameras mounted inside the car at the top corners of the windscreen. The result of the processing was fed to the driver through a set of output devices providing acoustical warnings, visual feedback, and steering wheel actuation. The ARGO prototype was used to test different ITS concepts: Manual Driving: the system monitored the driver activity; in case of potential dangerous situations it warned the driver with acoustic and optic signals Supervised Driving: in case of dangerous situations, the system took the control of the vehicle in order to keep it in a safe condition Automatic Driving: the system drove automatically, following the lane and localizing obstacles on the path; it was able to perform lane changes. Since the main challenge was the fielding of the vehicle in a real environment, a 1000 miles long test (the Mille Miglia in Automatico Tour) was carried out to assess the vehicle autonomous capabilities. During this test, ARGO drove itself autonomously along the Italian highway network, passing through flat areas and hilly regions including viaducts and tunnels, quickly varying road scenarios with changing weather conditions and intense traffic. The system demonstrated high robustness with respect to horizontal road signs, the presence of the guard-rail, forks, junctions, and highway exits, heavy 3

4 traffic conditions. Moreover, also high temperatures, different light conditions, and high speeds did not influence the stability and robustness of the whole system, both for hardware and software. The RAS project: The RAS (Surface Antarctic Robot) project was one of the first efforts towards the automation of driving functionalities in extreme environments: the vehicle developed within this project (a snowcat) was sent to the South Pole for automatic maneuvering during the Italian scientific missions. The aim of the project was the application of artificial vision to the autonomous driving of a platoon of snowcats for the transportation of people and goods in the Antarctica region. The final goal was to develop a vehicle able to automatically follow a leader vehicle. Since even small drifts from the driving path can be extremely dangerous due to ice cracks, vision was selected as the sensing capability able to deliver the highest performance in terms of precision in the localization of the tracks left by the previous vehicle. The Antarctica region presents extreme environmental conditions that make this application very challenging and different from the driving of unmanned vehicles on highways or urban roads. Several approaches were considered and specific filters were developed in order to cope with the problems typical of this environment, such as very low temperatures and possible snow storms, high brightness, low contrast, sharp shadows, direct sunlight and reflections, absence of ground reference or unknown terrain slope. The approach used to solve this artificial vision problem was suggested by the experience previously developed by the research group within the automotive field. In particular, a simple approach was preferred with respect to more sophisticated ones thanks to its easy implementation on the simple processing engine available on the snowcat. Following the approach that drove the development of these pioneer prototypes, alternative strategies were considered in order to evaluate different algorithmic solutions. The first solution investigated was based on traditional feature extraction techniques and exploited available libraries and previously developed experience. An innovative approach based on an evolutionary technique was also tested. The second solution confirmed the applicability to vision for vehicles of this emerging technique that still constitutes a hot research topic. The vehicle was tested both on a test site on the Italian Alps and in Antarctica in real environmental conditions. Track detection was successful in different situations even if noisy or critical conditions such as shadows, sun reflections, unknown terrain slope, and dark objects were present. As mentioned, at the time of the development of this projects the interest in the development of intelligent prototype vehicles was limited to a few far-seeing groups. For example the ARGO prototype was totally conceived, designed and implemented by VisLab, who internally financed the whole initiative. The RAS project, on the other hand, was funded by the national agency for the energy and the environment (ENEA, Ente per le Nuove tecnologie, l Energia e l Ambiente) who invested a large amount of money in innovative and risky research on unexplored topics. The choice of using only passive sensors to perceive the surroundings, both in the road and antarctic environments, was mainly dictated by considerations on their non invasiveness with respect to the environment. The cost of the sensor was also a main in concern in a project such as ARGO, which was oriented to the development of driving assistance systems that could be deployed in common road vehicles. In fact, in this kind of vehicles safety devices are to be considered as a cheap add-on. Another basic choice in these projects was the use of low cost standard processing hardware. The whole Mille Miglia in Automatico Tour was successfully completed with a 200 MHz Intel Pentium based processing system. The real-time implementation of vision-based driving assistance functionalities with a general purpose processing hardware was a significant result. We believe that these and other coeval similar projects helped the scientific community becoming aware of the possibility to perform vision-based automotive tasks in real-time with commercial hardware. The software systems developed for both these projects were stand-alone applications aimed at realizing specific driving functionalities. They were based on three main modules: data acquisition, image processing for feature detection (lane markings, obstacles, vehicles, tracks,...), output to devices for warning the driver or actuators. These applications are the core from which a very complex software framework evolved subsequently (see section III). The ARGO and RAS vehicles were two of the first prototypes with intelligent capabilities demonstrated onfield. Having to confront ourselves with the real world, we had to face a lot of problems typical of the outdoor environment (such as reflections, rain, road works,...) and of real vehicles (man-machine interface, interferences, temperature,...). One of the main achievements of the implementation and test in the real world was becoming aware of such problems and being able to solve some of them. 4

5 III. SUPERVISED SYSTEMS FOR THE AUTOMOTIVE INDUSTRY Following the outcome of the pre-competitive phase and the achievements of the pioneer projects, and thanks to the technological evolution of sensors and processors, the interest of the automotive industries towards Intelligent Vehicles began and gradually rose. We assisted and are still assisting to a flourishing of research initiatives oriented to the development of ADAS. The effort is mainly oriented to develop supervised systems rather than fully automatic vehicles, because of the high complexity of a full automation. Moreover, legal and liability issues have to be addressed as well if the human driver gets completely substituted by an automatic pilot. In addition, for a complete automation of civil transportation new infrastructures are required as well, and this entails a thorough redesign of mobility. Car manufacturers are pushing a lot in the ADAS direction, trying to early reach this new market. Moreover, the attention of the national and international governmental agencies towards these themes is very high nowadays: road safety, the protection of the environment, and the development of a sustainable mobility are felt as an unprocrastinable need. For instance, the EC in the esafety initiative launched in 2002 has set the objective of reducing the number of road fatalities by 50% within Another objective is the equipment of 20% of new cars with driving assistance systems. There are also efforts towards the regulation and standardization of such systems. Some supervised systems are already commercialized or close to the market. These systems address the most simple aspects of the driving task. They are based on the most elementary and/or settled sensing technologies and feature basic output (for example maneuver assistants, blind spot detectors, lane departure warnings, enhanced night vision). Conversely, for the most complex systems, such as the protection of vulnerable road users or stop and go driving, further development or assessment is needed. The research is oriented towards three main streams. (i) The first is the application of new emerging technologies that are becoming available at an affordable cost (such as infra-red cameras for night and day vision or laser scanners for robust all-weather obstacle detection). (ii) Another common trend is the use of multiple sensors with complementary characteristics and capabilities, and the fusion of their data to obtain a more reliable system. (iii) The third stream is the engineering of the systems that have been previously tested and demonstrated to be effective. The main challenge of the current research work is to attain robust and reliable prototype systems that can be transformed into products. The following paragraphs present some examples of research activities on prototypes based on different technologies developed by VisLab in collaboration with automotive partners and research centers. One of the research areas that appeared already promising at the beginning of the current decade is the protection of vulnerable road users. In this field VisLab has been carrying on a collaboration with the Volkswagen research center on the development of a prototype pedestrian detection system to be used for precrash or driver assistance applications. Thermal infrared is used as the main sensing device since it seems a promising sensor for pedestrian applications, but near infrared, along with specific illuminators, has also been investigated thanks to its minor cost and closeness to the market. To strengthen the system, the fusion with radar data has also been considered. Another automotive industry interested in the use of near infrared is the German company Hella, a large automotive supplier. The research we are carrying out for them concerns the use of headlamps and cameras working in the near-infrared domain for obstacle localization. The protection of vulnerable road users is a major theme also in the APALACI-PReVENT European project, aimed at contributing to road safety by developing and demonstrating preventive safety applications and technologies. In this framework VisLab is developing, together with Volvo, a start-inhibit system for heavy goods vehicles. Stereo vision is used to detect the presence of pedestrians or obstacles in the forward blind spot, so to warn the driver and also prevent the vehicle from taking off. As another example of different technologies applied, we are also collaborating within the same European project with the FIAT research center on the development of a system for road obstacle classification based on the fusion of radar and monocular vision. The system is able to classify vehicles, guardrails, and pedestrians. Automotive industries also show interest in the development of driving assistance applications based on color vision. In fact, even if color entails a larger quantity of raw data to be analyzed, nowadays DSP processing makes it a viable solution. As an example we are working for a large Italian automotive supplier, Magneti Marelli, on a new project regarding lane keeping and road signs recognition based on color vision. In the military field the leading edge technologies can be applied, even if they are more expensive. An example of the simultaneous applications of four cameras is the 5

6 work developed by Vislab in collaboration with the Vetronics Research Center of the U.S.Army TARDEC department. The project is aimed at robotic vehicles safety. Initially, a day light stereo vision system for the localization of human shapes was developed. The system has later evolved into a tetra-vision (4 cameras) system based on the simultaneous use of one far infra-red and one visible cameras stereo pairs. The main idea is to exploit the advantages of both far infra-red and visible cameras, trying at the same time to benefit from the use of each system. According to the general trend, in all these investigations and experimentations in order to obtain robust systems the fusion of multiple sensors was applied. Generally these projects share the need for a realtime software system able to acquire data from different sensors on the vehicle and to perform playback in laboratory for algorithm development. Anyway, each project needs specific data acquisition procedures. For this purpose we have developed a complex software framework. The applications implemented for the different driving functionality are integrated into this framework as plugins, and the system can be used both as a laboratory development environment for the algorithms, and as the ADAS software engine on the prototype vehicle. The system is featured by a number of acquisition modules tailored to different devices: the vehicle CAN bus, a variety of cameras, other sensors such as radars and laser scanners, network filesystem and disk acquisition for in laboratory post-processing. Multi-threaded processing is exploited to boost performance overlapping data acquisition, processing and output. Graphical software libraries and hardware acceleration are used to enhance the visual output. The great added value of this software framework is the capability to become an easy tool for the prototyping of many different applications in the automotive domain. IV. A NEW PUSH TOWARDS AUTONOMOUS INTELLIGENT VEHICLES: THE DARPA GRAND CHALLENGE While the first supervised systems are being deploying and are triggering a lot of interest, a new flourishing interest has turned out for autonomous vehicles. In fact, since the supervised systems are reaching a high level of maturity in the civil field, military organizations pushed again for the development of fully autonomous vehicles. As an example, the U.S. Department of Defense set the goal of making one third of ground military forces automated by In 2004 and 2005, the U.S. DARPA organized the Grand Challenge, a 200 miles long race for autonomous ground vehicles [7]. Similarly, in 2005, in Europe the German Federal Armed Forces organized the Land- Robot Trial to demonstrate the state of the art for autonomous or semi-autonomous vehicles. In both Grand Challenges, DARPA offered an appealing prize to the winner (1 million USD in 2004 and 2 millions in 2005) to boost the development and fielding of autonomous robotic vehicles. This winning idea demonstrated its efficacy since much of the development was autofinanced or externally sponsored. The investment of DARPA was paid off by the results achieved by many teams. More than 100 teams registered in 2004 while in the second race, 195 teams from all over the world entered the race. Moreover, an advantage of letting the teams free to chose their own sensing devices and technology, is the exploration of a wide number of solutions. This can be regarded as an evolutionary approach: different solutions are tested and the most effective approach wins. Due to the high complexity of such a task, reliability and the capability of dealing with extreme environments were mandatory (the Grand Challenges took place in the Mojave desert). The computing engines were developed with reliability in mind: they had to be able to continuously work without human intervention in an extreme environment. Particular care was used to develop systems able to work in high temperature and dusty conditions. Also sensors were selected using reliability constraints: they were generally installed outside vehicles and therefore they had to be placed in positions where they were less likely to be damaged by accidents or rocks during the race. Special cleaning devices were conceived to remove debris such as dust, water, and mud from sensors. Five prototypes on 195 successfully completed the race. The winning team (Stanley) reached the finish lane in 6:54 hours by using laser scanner and GPS sensors. H1ghlander and Sandstorm teams employed a similar approach and technology; beside the information coming from on-board sensors they also exploited a very precise prevision planning strategy. The Kat-5 vehicle used lasers as primary sensors as well. The TerraMax vehicle (an MTVR truck developed by Oshkosh Truck Corporation, VisLab, and Rockwell Collins, see box) on the other hand reached the finish line relying primarily on vision. The innovative approach used by the TerraMax team was the use of a trinocular vision system with different inter-camera distances; this system demonstrated to behave better than other approaches that relied on the use of moving camera heads. During the race, TerraMax reached a maximum speed of 68 km/h. This is impressive not only due to the size 6

7 and weight of this truck, but due to the fact that obstacle avoidance, path recognition, and trajectory planning were achieved at these speeds. Moreover, due to the race rules, TerraMax was paused during the night while the whole sensors suite and vehicle engine were were kept operational; the following morning TerraMax was allowed to move again and reach the finish line after 28 hours of continuous activity. This prolonged operational time demonstrated the robustness and reliability of the entire system. The DARPA Grand Challenge 2005 can be regarded as a true success, since five vehicles completed the race. As a result, DARPA already announced a further and even more challenging competition: the 2007 Urban Challenge. Fully autonomous vehicles have to complete a 60-mile long race in an urban environment in less than 6 hours obeying to road rules and negotiating traffic. [3] A. Broggi, M. Bertozzi, A. Fascioli, and G. Conte. Automatic Vehicle Guidance: the Experience of the ARGO Vehicle. World Scientific, Singapore, Apr ISBN [4] E. D. Dickmanns. The Development of Machine Vision for Road Vehicles in the Last Decade. In Procs. IEEE Intelligent Vehicles Symposium 2002, volume 1, pages , Paris, France, June [5] V. Graefe and K.-D. Kuhnert. Vision-based Autonomous Road Vehicles. In I. Masaki, editor, Vision-based Vehicle Guidance, pages Springer Verlag, [6] C. E. Thorpe, editor. Vision and Navigation. The Carnegie Mellon Navlab. Kluwer Academic Publishers, [7] U. S. DARPA. Grand Challenge Web Site. Available at V. A GLIMPSE TO THE NEAR FUTURE The Urban Challenge final goals, although promoted by a military oriented organization, are also strategic for the civil transport. When these goals will be met, the road to a technological transfer of these results to the car market will be open. VisLab will be again contributing to the worldwide research efforts by fielding a vision system for a new vehicle. These future technological achievements need to be supported by a completely new legislation that should take into account also liabilities and rules involving automatically driven vehicles. Standardization is another important keyword since UGVs, besides implementing standard driving tasks, will have to communicate among each other and will be part of a network including intelligent infrastructures. The next worlwide milestone, the 2007 Urban Challenge, although being a wide testbed for the latest technologies and the most promising data fusion approaches, will indeed provide a thoughtful estimation of the time span that we should wait before we sit on our own automatic car. In fact, the more positive results the Urban Challenge will deliver, the more the automotive industry will be stimulated in investing in automatic driving products. REFERENCES [1] M. Bertozzi and A. Broggi. Vision-based Vehicle Guidance. IEEE Computer, 30(7):49 55, July [2] M. Bertozzi, A. Broggi, M. Cellario, A. Fascioli, P. Lombardi, and M. Porta. Artificial Vision in Road Vehicles. Proceedings of the IEEE - Special issue on Technology and Tools for Visual Perception, 90(7): , July ISSN

8 The first vehicle prototypes that were tested and demonstrated automatic capabilities in real traffic and real scenarios. (This figure is intended for being displayed in a separate box). The Vamp vehicle prototype [4] developed by the University of Munich, Germany, and that was demonstrated in 1994 with a trip in automatic from Munich to Odense (Denmark). NAVLAB, developed by Carnegie Mellon University, travelled in automatic mode from coast to coast in the US back in 1995 (photo reprinted from [3]). ARGO was tested during a km trip along Italy in 1998: ARGO was able to safely keep inside the lane, detect obstacles, and -after a manual trigger- also overtake slower vehicles. 8