Abstract— Autonomous navigation in real roads has been a concern for several years now, especially for the Intelligent Transportation Systems community. Several interesting results have been obtained since the early 90’s, but the problem is so vast and manifold that even today it keeps many persons at academic and industrial levels engaged. These researchers are focused in the search for more, and more general, approaches for road and lane detection, traffic lights and many other perturbations on the normal flow of perception from the road and its entourage. Perception and navigation are the main two components of this huge problem, where the second depends largely on the performance of the first. In Portugal, a competition for Autonomous Driving was created in 2001. It intended to promote the development of solutions for these perception and navigation problems. Currently, the competition accounts for several challenges and a University project named ATLAS has developed a series of robots that have outperformed all the competitors for five years in a row. The paper presents the main techniques and algorithms that lead to this success and formulates the bases to migrate the solutions to real road conditions. Perception of road and its features are given the main focus. I. INTRODUCTION UTONOMOUS navigation of cars and other vehicles on real roads is perhaps a dream as old as the modern automotive industry. Although many interesting results have emerged in the last 15 years, no definitive answer exists in terms of robust perception and even less in safe and robust driving. To promote developments and contributions to near the reality to that dream, several groups have devised many sorts of activities among students and research communities. One of such groups proposed in Portugal, in the late 90’s, a competition 0 for autonomous robots where some sort of road plunged with obstacles and other perturbations was to be traversed intelligently by self-contained machines. The Portuguese Robotics Open (ROBOTICA) was then created and the competition of Autonomous Driving is now its oldest competition, among others created meanwhile. The main goal of the competition is to complete two rounds of an 8-shaped path simulating a real road (Figure 1), although with some controlled parameters such as good line definition on the floor. Nonetheless, the challenge is quite demanding when details are observed closely. In its last level M. Oliveira is a PhD Student at University of Aveiro. The author thanks FCT for funding under the grant SFRH/43203/2008. Email: mriem@ua.pt. V. Santos is with the Department of Mechanical Engineering and the Centre for Mechanical Technology and Automation of the University of Aveiro. Email: vitor@ua.pt. of complexity, the competition comprises a zebra crossing area defining precise stopping areas in case the traffic lights, suspended above it, force the stopping. Random generation of traffic signals also with random values for their duration make the problem demanding for entry level competitors. That is however negligible when compared to a tunnel where light conditions change dramatically and when the robots reach the road maintenance area where an alternative road delimiter is used simulating temporary path detours; in that case, the orange and white stripes and cones override the normal road white lines. Even more challenging is to have, above all this, obstacles in unknown positions and, finally, at the end of the trial, complete the challenge with the pièce de resistance which is a compelling parking area located somewhere outside the road track having two places to park but one of them randomly occupied by another obstacle. The entire challenge requires abundant vision systems and other techniques both for perception and navigation. Very few robots in the ten editions of the competition have managed to complete all the challenges. Being able to cope with these many challenges certainly is an indicator that real roads in real environments can also be dealt with. The task with real roads is obviously more difficult, but the problems share principles and methods to reach the solution. Parking area with an unknown obstacle Unknown obstacle Tunnel Road maintenance area Zebra crossing and traffic light panel Figure 1 – Model of the autonomous driving competition environment. The remainder of the paper introduces some related work in road perception, and then continues with the techniques used by the ATLAS robots, especially in road segmentation and its embedded obstacles and accessories. Conclusions and perspectives for future work are drawn at the end. Since 2004, in seven editions, the ATLAS robots have obtained, 2 third places, 3 second places, and 5 first places in a row in the Portuguese national competition. II. PREVIOUS WORK ON ROAD PERCEPTION Several approaches have been attempted in order to solve the problem of extracting the road using visual information only. There are two separate threads in what road detection Autonomous Driving Competition: Perception Approaches used in the ATLAS Project M. Oliveira, V. M. Santos, Member, IEEE A