Traffic jam driving with NMV avoidance
Mechanical systems and signal processing
In recent years, the development of advanced driver assistance systems (ADAS) -mainly based on lidar and cameras -has considerably improved the safety of driving in urban environments. These systems provide warning signals for the driver in the case that any unexpected traffic circumstance is detected. The next step is to develop systems capable not only of warning the driver but also of taking over control of the car to avoid a potential collision. In the present communication, a system
... of autonomously avoiding collisions in traffic jam situations is presented. First, a perception system was developed for urban situations-in which not only vehicles have to be considered, but also pedestrians and other non-motor-vehicles (NMV). It comprises a differential global positioning system (DGPS) and wireless communication for vehicle detection, and an ultrasound sensor for NMV detection. Then, the vehicle's actuatorsbrake and throttle pedals -were modified to permit autonomous control. Finally, a fuzzy logic controller was implemented capable of analyzing the information provided by the perception system and of sending control commands to the vehicle's actuators so as to avoid accidents. The feasibility of the integrated system was tested by mounting it in a commercial vehicle, with the results being encouraging. problem is the driver's difficulty in reacting sufficiently quickly to unexpected circumstances. Two of the hottest topics in the transportation field correspond to driving in urban environments-following a leading vehicle at low speeds and reacting quickly to NMVs. The solution will need to use sensory information to autonomously manage the actuators-i.e., the throttle and brake pedals. One of the most challenging design issues in the short-to-medium term is therefore to develop such a sensor system for urban traffic conditions. In this sense, Volvo has developed the City Safety system for its vehicles. This system helps driver to entirely avoid a collision -when vehicle speed is under 15 km/h -or reduce as much as possible the speed prior crashing  . Moreover, Volvo has recently implemented a novel solution called Collision Warning with Auto Brake (CWAB) which helps drivers to avoid collisions at higher speeds-up to 30 km/h. The system merges the information of a long-range radar with a forward-sensing wide-angle camera to monitor the area in front of vehicle  . Four principal types of sensor technology have been used in approaches to solving the NMV detection problem: laser radar (lidar), microwave radar, infra-red thermal imaging, and machine vision. While under optimal conditions lidars have high accuracy, wide angle coverage, and precise target location, their performance is lowered in inclement weather and/or when dirt collects on the lenses . Most radars use some form of beam scanning to resolve whether targets are in the same roadway lane, an adjacent lane, or an oncoming lane. While microwave radars are able to operate accurately in inclement weather, they have a narrow field of vision and are prohibitively expensive for mass-produced cars  . The images from a video camera are usually produced at a 25 Hz frame rate, and these frames can be stored and processed by a computer as a matrix of pixels. Image processing techniques are used to provide information about the presence of obstacles up to 50 m away  . However, as humans vary significantly in size and shape, there are difficulties in obtaining robust results with image analysis techniques, especially in crowded areas. Moreover, extreme lighting conditions (day/night) can dramatically reduce the effectiveness of the detection algorithms. While passive (thermal) infra-red imaging already provides night vision in commercial ADAS systems , the information it provides is neither trustworthy nor persistent enough to be used in automated cars. Obstacle detection at short distances is usually approached using microwave radar, radio-frequency capacitance, infrared multi-beams, or ultrasound. Capacitance sensors measure the change in dielectric constant that occurs when an object whose dielectric properties are different from air approaches the sensor. They have yet to be implemented in massproduced vehicles because they are highly sensitive to radio-frequency interference. Ultrasound technology is widely used in production for parking man?uvres because it offers wide-area, near-distance beam coverage, and is low-cost  . Ultrasound sensors can be used in air at a range of 15 m with simple and cheap electronics. Their main advantage over other types of sensor is that they can detect any NMV in the roadway, which makes them well suited to preventing, or at least minimizing, damage. Their main disadvantage is that they are influenced by turbulence caused by the wind which disperses the ultrasound waves or produces false echoes. The present communication is part of a collaboration between the AUTOPIA program at the Centre for Automation and Robotics (CAR, CSIC-UPM) and Santander University. The goal is to develop an approach to ACC plus NMV detection in urban environments. To this end, a prototype vehicle equipped with a DGPS combined with an Inertial Measurement Unit (IMU) to obtain the vehicle's positioning with high accuracy has been instrumented with automatic driving capabilities. Using wireless communications, the vehicle can receive data on the position of the other vehicles in its vicinity to perform ACC man?uvres. An ultrasound sensor detects obstacles on the road. With these two systems, measurements can be made of the distance to any object located in front of the vehicle. If the distances determined by the GPS and the ultrasound sensor are equal, then a leading vehicle has been detected. If the distances are different, it is assumed that an NMV has been detected between the two motor vehicles. In a later phase of the research program -the control stage -a fuzzy logic controller is designed to act autonomously on the throttle and brake pedals. The rest of the paper is structured as follows. Section 2 describes the system developed to control the vehicle, and Sections 3 and 4 present the sensor systems used to acquire environmental information on any leading vehicle and NMVs, respectively. Section 5 presents the fuzzy-based control algorithm, and Section 6 the brake and throttle automation system. Experimental results of trials on a private driving circuit are described in Section 7, and some concluding remarks are given in Section 8. The on-board system The control system is divided into three stages (see Fig. 1 ). The first stage is in charge of acquiring all the sensorial inputs from the environment. The vehicle is equipped with a differential global positioning system (DGPS) located close to the rear-end of the car, and an inertial measurement unit (IMU) located as close as possible to the centre of gravity. They are in charge of positioning the vehicle in the road  . The commercial Citroen C3's controller area network (CAN) bus is available to acquire information on the vehicle. For this application, only information coming from the speedometer is used. An ultrasound sensor is located on the front bonnet of the car. A Personal Computer Memory Card International Association (PCMCIA) Proxim Wireless ComboCard is used to obtain information about other vehicles in the vicinity. Sections 3 and 4 will describe the sensors used to carry out autonomous ACC with NMV detection, respectively. The second stage is in charge of selecting and executing the best control strategy in each driving situation. To this end, different fuzzy controllers [4,    have been developed to take into account various traffic situations. The fuzzy controller implemented in the proposed driving aid system will be described in Section 5.