Ideally an intelligent autonomous car would have an automatic pilot, which can park itself and guide the vehicle through dense traffic in towns and at high speeds between towns. At any time the driver would have the ability to switch between automatic and driver control. During periods of manual control the system would act in an advisory capacity warning the driver of hazards and giving information about route guidance or traffic congestion.
Such a vehicle would retain the convenience and fun of private transport but would take the drudgery out of driving during automatic operation. Such ambitious goals are being seriously considered by vehicle manufactures. Fully autonomous systems will not appear for some time although they may be introduced sooner if the traffic environment is greatly constrained. More viable for medium term are semi autonomous systems where driver maintains responsibility for the overall control of the car. Thus they will provide platform for providing the technology for later fully autonomous systems.
This paper focuses on some of the semi autonomous systems that are currently being investigated in Europe, US and Japan which include:
Â¢ Autonomous intelligent cruise control systems.
Â¢ Lane support systems.
Â¢ Collision avoidance systems.
AUTONOMOUS INTELLIGENT CRUISE CONTROL SYSTEM
Traditional cruise control systems have been in use for many years. They maintain a constant vehicle speed, set by the driver, thereby improving comfort in steady traffic conditions. In congested traffic conditions when speeds vary widely these systems are no longer effective. The use of cruise control would be significantly increased if the vehicle speed could automatically adapt to the traffic flow.
Fig-1: Architecture of Autonomous Intelligent Cruise Control System
The Autonomous Intelligent Cruise Control (ACC) systems add $1500 to $3000 to the cost of a car, use laser beams or radar to measure the distance from the vehicle they are in to the car ahead and its speed relative to theirs. If a car crosses into the lane ahead, say, and the distance is now less than the preset minimum (typically a 1- or 2-second interval of separation), the system applies the brakes, slowing the car with a maximum deceleration of 3.5 m/s2 until it is following at the desired distance. If the leading car speeds up or moves out of the lane, the system opens the throttle until the trailing car has returned to the cruise control speed set by the driver.
The addition of a radar sensor to the front of a vehicle would provide the necessary range and velocity information for this task. Automatic control of brakes and throttle would allow the longitudinal controller to maintain a constant time interval behind the vehicle in front. Such systems are commonly referred to as Autonomous Intelligent Cruise Control Systems (AICC) and aim to improve driver comfort and conveniences. It is important to remember that these systems are quite different from the fully autonomous car. The driver is always responsible for driving and must deal with emergency situations. The system is only capable for fine longitudinal control and not emergency braking. The system is further enhanced by allowing operation down to zero speed which would increase comfort during traffic jams.
Fundamental to any AICC system is a sensor that can reliably detect obstacles in the traffic environment in a variety of conditions. Microwave radar is a method for detecting the position and velocity of a distant object. A beam of electromagnetic radiation, with a wavelength between 30cm and 1mm is transmitted and reflected back to the transmitter by the object. Velocity and Range can be derived by measuring the Doppler frequency shift and time of flight of the transmission. A major advantage of microwave radar is that the performance is not affected by the time of the day, and therefore no driver adaptation is required for night- time driving. The performance advantages of radar over other sensors are enhanced during poor weather conditions. Systems that rely on visible light are known to suffer significantly in the very conditions for which they are relied upon the most. Experience of microwave radar operation has shown that reliable results can be obtained, even in inclement weather conditions.
The operation of microwave radar falls broadly into two categories: Pulse and Frequency Modulated Wave (FMCV). Pulse systems rely on measuring the time of flight of a pulse that is proportional to range.
Information from the sensors goes to the Vehicle Application Controller (VAC), the system's computing and communication center. The VAC reads the settings the driver has selected and figures out such things as how fast the car should go to maintain the proper distance from cars ahead and when the car should release the throttle or downshift to slow down. Then it communicates that information to devices that control the engine and the transmission.
Applications in Cars
In May 1998, Toyota became the first to introduce an AICC system on a production vehicle when it unveiled a laser-based system for its Progres compact luxury sedan, which it sold in Japan. Then Nissan followed suit with a radar-based system, in the company's Cima 41LV-2, a luxury sedan also sold only in Japan. In September 1999, Jaguar began offering an ACC for its XKR coupes and convertibles sold in Germany and Britain. Mercedes' system is an option on its C-Class and S-Class models, which are available in Europe; it was developed by M/A-Com, Lowell, Mass., and uses a radar made by Filtran Microcircuits Inc., in West Caldwell, N.J.
LANE SUPPORT SYSTEM
A recent US study shows that 25% of all accidents are caused by unintended lane departures. Mounting evidence suggests that drivers falling asleep, although it is difficult to prove, as many drivers will not admit to falling asleep,talking to a cell phone,disruption of children etc. cause many of these accidents.
To address this, Valeo a European maker of automotive switches and sensors, and Iteris, U.S. developer of intelligent transportation system technologies, have produced AutoVueâ€an embedded camera-based lane-marking recognition and warning system. By monitoring the visual lane markings on the road and signaling the driver with an audible or tactile warning, AutoVue alerts a tired, distracted, or inattentive driver that he or she is about to leave their lane
The lane support system audibly warns the driver of unintended lane departure. Audio messages are issued as the driver crosses a lane marking without using the indicators and the balance of the stereo sound system is controlled to denote the direction of road departure. The lane markings are detected by processing images from a video camera. The current system achieves lane-marking detection at 15 frames per second. Road edges with no lane-markings are detected although they are less reliable. Preliminary studies show that it is possible for positions. The frequency of oscillation as seen in adjoining figure appears to be about 1/8 hertz for normal alert drivers. This information could be used for determining the state of driver alertness, although further investigations are necessary before this can be proved.
More effective than audible warnings, that could irritate the driver, is the use of haptic feedback in the steering wheel. The system could provide position related assistance/resistance in the steering wheel. The system could provide an artificial feel of road camber on either side of the lane. The driver is still expected to steer the vehicle but experiences the sensation of driving along the bottom of a bathtub. Again the system is deactivated by either the use of the indicators or by the driver exceeding a torque/motion input threshold at the steering effort, improved steering stability and safety.
Fig-4: Architecture of Lane Support System
The key components of the lane support system are shown in the adjoining figure. A video camera and processing measures the lane markings. Another sensor measures the torque that the driver is applying to the steering wheel. The Electronic Control Unit (ECU) takes this measurement and drives a motor, which in turn applies torque through a gearbox to the steering mechanism there by assisting the driver. The incorporation of clutch improves the feel (by disengaging the motor) when no assistance is required and
also improves fault tolerance.
When AutoVue detects an unplanned lane departure, a sound is generated to alert the driver. The sound can be directional so that a left-hand departure will result in the sound being generated in a left vehicle speaker and a right hand departure in a right speaker. A camera views the road through a "wiped" section of the windshield, near the vehicle center line as shown below.
The image that follows shows the outline of the view processed in the crosshatched area seen in the previous diagram. The system camera is mounted on the windshield inside the cabin, near the center line of the vehicle. The imager is mounted high on the windshield for better functioning of the algorithm to track lane markings and to accommodate viewing over the hood.
Within the system is a central processor unit running a program that performs the lane-tracking and warning algorithm. The CPU runs this algorithm in conjunction with other embedded environment tasks (communications, initialization, and general task management). The lane-tracking algorithm accepts image input from the system camera and examines the lane markings. It determines whether or not a lane departure is imminent.
Upon such determination, and if all logic permits (for example, a turn signal is not active, and no system errors are present), a warning will be issued in association with the departure. If the system is also connected to a Controller Area Network (CAN) a message will be generated indicating which side of the vehicle the departure is occurring. This message can be used to create a warning to the driver through interfaces such as the infotainment system or steering system. Audible warnings are designed to be loud enough to overcome wind noise from an open window, as well as engine, traffic, and other noises, without being annoying.
The AutoVue module hosts a complete lane tracking software solution including image capture and improvement, lane tracking algorithms, and drivers for the various system interfaces.
Lane Departure systems have received positive feedback from commercial vehicle operators. A driver satisfaction study of Iteris' Lane Departure system was conducted in 2004 including responses from 140 drivers in the U.S. and 100 in Europe that have used the Lane Departure system. A brief summary of the U.S. data from the study is as follows:
Â¢ 98% believe the system can prevent accidents
Â¢ 92% believe the system is a valuable safety feature
Â¢ 71% say the system has made them safer drivers
Â¢ 80% normally drive with the system enabled
Â¢ 97% are satisfied or very satisfied with the system
Implementation in cars
Already, Nissan's luxury division, Infiniti, is offering a system, developed by Iteris and supplied by Valeo, on its 2005 FX45 sport utility vehicle, Infiniti's second application of the Valeo LaneVue system will be on its 2006 M performance sedan, which goes on sale next spring.
In Europe, Valeo is also supplying the system to the 2005 C4 and C5 sedans from Citroen, recently introduced at the Paris auto show.
COLLISION AVOIDANCE SYSTEM
Every year 50,000 people are killed on the roads in the European countries. The figure for the USA is similar. In Britain out of 240,000 accidents in a typical year there are 5,000 fatalities and 64,000 injuries. Many safety innovations in the areas of banking systems, airbags, body structures, steering and suspension have already had a beneficial effect. However, accident frequency and severity have still remained unacceptably high. Accident studies places most of the blame on drivers. Automatic Collision Avoidance Systems could have many advantages over human drivers; they donâ„¢t get tired or distracted, they can simultaneously monitor all sides of the vehicle, they generally have faster reaction times and they donâ„¢t panic. Apart from human tragedy these systems could significantly decrease the economic cost of road accidents and reduce the inconvenience associated with the resulting traffic jams.
The main requirement for a collision avoidance system is to be able to predict accurately the likelihood of an imminent collision. If a collision is likely then the system should respond in such a way as to reduce the threat. The response could involve the automatic control of the vehicle or simply an appropriate warning to the driver. Crucial to the success of any avoidance system is the ability to reliably identify and locate different obstacles in the complex and unconstrained traffic scene, where conditions can vary from thick fog at night to bright glare from the sun on a clear day.
Collision Avoidance Sensing
Collisions can occur at any point on the car, therefore sensor coverage should ideally extend to 360 deg. It is probable that no sensor can satisfy these strict demands. Headway detection systems have received most attention as they potentially cover majority of accident situations. However they are difficult to implement because of the large maximum range required for driving at speed. Many false alarms are generated by radar systems illuminating objects on the bends that are safely off the vehicle path. False alarm rate should be very low in a collision avoidance system to avoid driver irritation. If there is automatic braking then they should be eliminated. Calculating collision paths on long ranges and on bends requires a detection system with wide field view, a high angular resolution and the detection of the road geometry.
Fig-6: Architecture of Collision Warning System
Following sensing devices are generally used:
Â¢ Front sensing system (MMW Radar, LASER Radar)
Â¢ Side Sensors
Â¢ Vision Sensors (CCD Video Camera)
Â¢ Host Vehicle Sensors
Information from these sensing systems flow to the collision warning processing module eventually to the Driver Vehicle Interface (DVI) which provides the appropriate warning cues to the driver. Each of the sensing systems receives information from host vehicle state which includes:
Â¢ Differential wheel speed
Â¢ Vehicle speed
Â¢ Steering angle
Â¢ Yawing rate
The date is then send to collision warning processing module. The data contains information like lane path, vehicle speed, range etc, relative to host vehicle. The Collision Warning Processing Module will then combine data from the active sensing systems and passive sensing systems to accomplish object detection, target tracking, in path target identification and Threat Assessment. Threat Assessment falls in three categories:
Â¢ Time To Collision (TTC)
Â¢ Time To Avoidance (TTA)
Â¢ Threat or No Threat Decision
If the identified detected target is accessed as being a potential hazard to the host vehicle then appropriate warning cues will be issued.
The current Collision Avoidance System uses a headway video camera and a CCD device. The radar system is similar to AICC radar system. The video system detects cars and lane markings up to a distance of 60m. The low-level image processing acquires images from a miniature CCD video camera situated by the rear view mirror. Edges are extracted using a technique called Intelligent Thresholding. The edges are then thinned using morphological techniques. The thinned edges in the image are then traced by fitting straight lines to them and then turned into line vectors, which are then categorized on the basis of angle. The horizontal lines are used by car detection algorithm. Range is estimated from the road to camera geometry by assuming a flat Earth. As, a result the range data are quite noisy due car pitching on hilly roads and small possible variations in the determination of the car position in the image plane.
A very high computational rate is required for image processing. Hence microprocessors with high computational rate are required.
Data fusion is a collection of techniques for combining the measurements from more than one sensor to provide a more unified result. The sensors used can be of the same or different type. This has several benefits:
Â¢ The overall estimates of parameters can be more accurate than for individual sensor estimate, as they reinforce each other.
Â¢ Any parameter need not depend on one sensor alone. This has benefits of Fault Tolerance by allowing redundancy to be introduced into the system. For example, if the system had infra red and video based sensors, it could survive the failure of either of these and continue to function although accuracy and performance in certain conditions can be reduced.
In Lucas system data fusion can gives major benefits to object detection as the two sensors could be used to complement each other extremely well. The video produces good lateral image but it is not able to provide good estimates of range. The radar conversely produces good estimates of range and hence good relative velocity estimates, but has poor lateral positional accuracy. Thus by fusing data from these two sensors the object position can be localized to a better accuracy by considering the intersection of the two areas of positional uncertainty generated by each sensor.
Another possible benefit of data fusion is that of object identification, by combining the expected responses of an object in the sensors. For example, the image processing may confuse a stationery pedestrian and a traffic sign, especially at longer ranges, as they both are tall, thin objects. However fusing data would allow an unambiguous decision to be reached immediately as the two objects have radically different radar responses.
Parameter Radar response
Pedestrian Tall and thin Poor
Traffic sign Tall and thin Good
Fig-8: Object Identification by Data Fusion
Having detected a hazard the system must respond in the most appropriate way so as to avoid the hazard. Evidence suggests that a small reduction in driver reaction time will dramatically reduce the number of accidents. This can achieved either by intervening with the controls of the vehicle or by warning the driver. Direct intervention will have the most significant result but causes the greatest liability concern. In the worse case the car would contravene the driverâ„¢s wishes by emergency braking just as the driver decides to brake and overtake. Driver warning is the safer option but there are still drawbacks. In critical situation there will be no time to warn the driver or the warning might even distract the driver from hazard. If there are too many warnings, false alarms, or if the driver is already aware of hazard then warning will be irritating.
Usually, warnings are either visual or audible. The visual warnings must be in the driverâ„¢s field of view at the time of warning. Audible warnings are generally more obtrusive and hence irritating if inappropriate. However also effective are haptic or tactile warnings. Haptic feedback in the steering mechanism has already been described earlier in lane support system, but the same system can be used to increase the steering reactance if there is an obstacle in the driverâ„¢s blind spot during a lane change maneuver. Kinesthetic warnings invoke a very quick response by jerking the car or by moving the seat. The driver rapidly decelerating the car can quickly wake even sleeping passengers. If the jerk situation is short then there will be little effect on the carâ„¢s speed leaving all options open for the driver.
Detection of hazards can be calculated by considering the trajectories of other objects in relation to your vehicles, its operating envelope and the state and intentions of the driver. If the driver is inattentive then action should be taken, but if the driver is alert then there is no need to take any action. Hence the decision process should include state and intentions of the driver. Monitoring the driver is a hard problem; it is difficult to distinguish between a genuine lane change and a lane drift due to drowsiness. Progress has been made using neural networks to learn the driver behavior. A backward error propagation network can predict if the driver is going to overtake or brake with a success rate of 90% when trained on only 15% of total input data.
In the situation that require direct of the vehicle, recent work has shown the use of neural networks for emulating driver brake and throttle control. Such a system should be particularly useful for AICC where drivers feel most comfortable, when the system mimics their own driving style.
Â¢ Maintains safe, comfortable distance between vehicles without driver interventions.
Â¢ Maintains consistent performance in poor visibility conditions.
Â¢ Maintains continuous performance during road turns and elevation changes.
Â¢ Alerts drivers by way of automatic braking. Â¢ Minimizes speed differentials between vehicles.
Â¢ Reduces throttle and brake management.
Â¢ to be able to adapt - therefore, learning ability
Â¢ to be autonomous - therefore, ability to sense, model and provide output
Â¢ the need for more safety and convenience in the vehicle
Â¢ to overcome the loss of life and financial cost
This paper has described a wide variety of automotive applications for intelligent autonomous systems. The fully autonomous car is probably not viable in the foreseeable future. Semiautonomous systems as discussed above are technologically feasible but issues such as driver acceptance, reliability, safety and product liability have yet to be resolved. AICC and Lane Support Systems not only reduce driver comfort but also reduce the risk of an accident. Collision Warning System will be useful for alerting a distracted driver to hazard provided the time to impact is not too critical.
When considering the benefits, cost is a major consideration in the automotive market. Much of the technology tends to be expensive as it originates from low volume/high cost military markets where volumes are high such as microwave satellite receivers and door openers then cost have become less and markets have been successfully exploited.
The end objective is to be able to survive in an evolving environment and changing circumstances, a representation - of real-world system.
1. Intelligent Autonomous systems for cars â€œ By R.H Tribe.
2. Autonomous Intelligent Cruise Control with automatic braking â€œ By Martin.p, SAE, Detroit, February 1993.
5. Automotive Design Line