As humans, object recognition is part of everyday life, which is based on our senses. Computerized object recognition is the future of automobiles. To go from human object recognition to computerized object recognition is a huge step. Every year thousands of road accidents take place worldwide, mainly because of human error. To mitigate this, automobiles can be made completely automated, not requiring human intervention at all. Autonomous cars also bring about advantages as in fuel efficiency, comfort, and convenience, thus leading to vast research, worldwide. One key factor to achieve success in this field is creating better obstacle detecting sensors, and Artificial Intelligence (AI) paves way for incorporating this.

Artificial Intelligence is used by a computer, in the same manner, humans use their intelligence. This is very much helpful in object/obstacle detection, cruise control, and also in navigation. The various AI techniques used are fuzzy logic and Artificial Neural Network (ANN)

In the autonomous car now-a-days, many passive and active sensors can be predominantly used like cameras, laser sensors, radars, ultrasonic sensors, and GPS sensors. LIDAR sensors illuminate the target with laser light. This light gets reflected and the time delay or interval is measured. This helps in creating 3-D representations of the target. The position sensors provide the exact location of the car on the map for navigational purposes by providing details of altitude, latitude, and longitude. Multi-axis sensors are commonly used. Radio Detection And Ranging (RADAR) uses microwaves or radio waves for detecting the direction, distance, and also the speed of an obstacle in the path of the autonomous car. The waves get bounced off or reflected and this reflected wave is detected in the same site of the transmitter.

Rathod (2013) defined an autonomous vehicle as a passenger vehicle that drives by itself without any human intervention. This vehicle is known as an autopilot, driverless car, auto-drive car, or automated guided vehicle.

Rouf, Ali, and Hussain (2018) explained AI, which is a discipline of the computer system responsible for analysing various visual-inputs, such as facial, object, and gesture recognition. Hence AI is used in all autonomous vehicles.

Sun, Bebis, and Miller (2004) defined activities such as detecting the distance of an object by measuring the travel time of a signal emitted by the sensors and reflected by the object, it includes lasers, LIDAR. Research depicts that these sensors have a low spatial resolution, slow scanning speed, and interference among sensors at the heavy traffic. Sun et al., suggested that sensors need to be improved by using the neural network and fuzzy logic technology to enhance the performance of sensors.

Goerick, Noll, and Werner (1996) described the feature of car detection and tracking, CARTRACK system, it is a specialized monocular visual sensor system for detecting, tracking, and measuring rear or front views of automobiles in image sequences taken from the viewpoint of a car. The system consists of pre-processing, classification, or detection modules. Here, this classification and detection task is performed by means of Artificial Neural Network (ANN). The main functions of ANNs are detection, identification, classification, localization, and prediction of object movement. The speed of the Artificial Neural Network has gained by the pre-processing method as well as the integral treatment of image regions. Goerick et al. (1996) explained the different pre-processing methods, where grey-scale images are pre-processed by a method called as Local Orientation Coding (LOC). The image features obtained are bit strings, each representing a binary code for the directional grey-level variation in the pixel's neighbourhood. Grey level variations are the illumination conditions that vary, object tilting, occlusions, differently resolved structures depending on the distance of the object which will be under consideration, noise and perturbations induced by the recording and processing equipment, different viewpoints and also the type of cars, that is cars that differ in shape, size, and colour.

Naranjo, González, Reviejo, García, and De Pedro (2003) stated that fuzzy control deals with the input and output variables, here the knowledge of any mathematical model of the processes involved is not at all required. The relationship between input and output variables is expressed in sentences that represent or mimic human thinking and approximation, the variables, specifically the output variables, are normalized and their values can be applied to low-level controllers easily. These low-level controllers act directly upon the physical actuators. These are discussed in detail in this paper.

Machine learning is a field of artificial intelligence that uses statistical techniques to give computer systems the ability to "learn". Autonomous vehicles make use of deep learning and artificial intelligence to make informed decisions and detect the surrounding environment, just like a human being.

Hofmann, Neukart, and Bäck (2017) studied about machine learning, which includes both supervised and unsupervised learning methods. Research reveals that machine learning helps to learn, understand, and also to interpret information, behave adaptively, plan, make inferences, solve various problems, think abstractly. Machine learning helps to understand and interpret ideas and language which will help Artificial Intelligence application in an autonomous car.

Supervised Learning Algorithms helps in detecting variables, such as traffic signals; images of soil changes, light intensity, noising or blurring in the data, etc. However, Supervised Learning Algorithms will give same results irrespective of weather changes.

Unsupervised Learning Algorithms will characterize a data set in general, it is used to group the data set to identify the relationships between individual data points and group them into clusters. Unsupervised machine learning process is efficient in the detection of an object, traffic signals, and images without human interventions irrespective of weather conditions.

In driverless vehicle, control layer is made up of the fuzzy control system and knowledge-based system. The syntax for fuzzy rules is: IF X [OR/AND Z] THEN Y, where X and Z are fuzzy input variables and Y is the fuzzy output variable.

Chen, Zhang, Wang, and Wang (2003) identified the various disadvantages of Intelligent Transport Systems (ITS), such as people-centric, privacy, social security, human health, etc. The author suggested that to overcome this, Data-Driven Intelligent Transport Systems (DDITS) would be a more suitable method.

Data-Driven Intelligent Transportation System supported by a large amount of data that are collected from various resources allows to use interactively pertaining to the transportation system in, convenient and reliable ways to improve the performance of transportation systems.

An autonomous vehicle needs to detect the appropriate path to its target, which is the reason it is equipped with a navigation system. Also, the vehicle should be able to know its surroundings and should acquire knowledge about the surrounding, hence it is equipped with a sensory system.

Cavaretta, Chou, and Madani (2005); Greitzer (2005); Martin (2005) have presented different approaches to handling uncertainty by the theory of fuzzy sets, which allows great flexibility in the treatment of information.

The ultrasonic sensor is composed by two fundamental elements such as the acoustic transducer and a ranging circuit board as per Poloni, Ulivi, and Vendittelli (1995). Research also stated that the distance of the obstacle is calculated instantly by finding the time delay between the transmission and the reception of the ultrasonic waves. More sophisticated techniques could be used, including real-time signal processing, the limitations of this method is it requires expensive computation devices. Hence fuzzy theory has been used to overcome certain uncertainties of typical sensors.

It is noted that the first commercial application of fuzzy logic for speed control and continuously variable transmission dated back to 1988.

The human knowledge of the control behaviour is used by fuzzy logic controllers. The fuzzy logic control is defined by the collection of rules. All these rules will be in the form of 'If – Then' statements. Crisp sensor inputs are spontaneously converted to the fuzzy variables, which are processed in opposition to the rule base. The crisp control value is obtained by converting back the combined result. A few notable applications are used in Anti-lock Braking Systems (ABS), engine control, anti-skid, and climate control systems. Syed, Filev, and Ying (2007) proposed an application of fuzzy logic in a hybrid electric vehicle control system. They used the 'gain scheduling algorithm'. The main role of the fuzzy logic components is to accordingly adjust the Proportional Integral (PI) to different operating conditions.

The direct measurement of feedback data from the system parameters is impractical. Reasons being complexity, dynamic nature of the system, and noise. Marko, James, Feldkamp, Puskorius, and Feldkamp (1996) demonstrated that neural networks could be trained to mimic “virtual”, ideal sensors which improve the diagnostic information from existing sensors on production vehicles. The most famous application of neural – network - based sensors is the online diagnostics of engine combustion failures, featured in the Aston Martin DB9 engine control system.

A fact that engine misfires lead to excess vehicle emissions and fuel consumption, improves the application of neural - network - based virtual sensors. Neural networks, trained artificially by introducing failure in combustion can detect a misfire with a very high level of accuracy, based on information like load, engine speed, and phase of the cylinder firing sequence.

Lane departure warning, adaptive cruise control is some of the intelligent vehicle technologies in use today in almost all the vehicles. Examples of intelligent vehicle technologies existing today include lane departure warning, adaptive cruise control, and the parallel parking assistant.

The analysis of images from video cameras uses AI techniques like machine vision and pattern recognition. The fusion of the non-continuous data from various sources would benefit from the application of neural networks in the same way the virtual sensors are used in engine control. The implementation of a real-time response to the changes in driving conditions may take advantage of fuzzy logic. Miller and Tascillo (2005) describe the prototype of a system that identifies and classifies objects in close proximity range using a neural net approach to select the best path to avoid an accident.

Naranjo, Sotelo, García, and De Pedro (2007) discussed the application of a fuzzy co-processor to the autonomous vehicle. Fuzzy control is none specialized. The fuzzy control does not require any mathematical model knowledge, but it deals only with input and output variables. Human thinking is very closely incorporated by establishing a relationship between input and output variables to the sentences, which represent human thinking and decisions made by humans.

The values of output variables can be applied to low-level controllers very easily, as per the research. In the case of the authors' model, if a car was not subjected to any changes, then only low level were to be modified when changes in the dynamic environment took place.

To follow human driving, the authors represented human approximate reasoning. Fuzzy reasoning is defined as the process or processes by which a possibly imprecise conclusion is deduced from a collection of inexact premises. Next, the authors determined the fuzzy values, based on the car's parameters. Simulators were used to incorporate this. Lastly, two mass-produced cars were instrumented by them in order to allow automatic fuzzy control on the steering and the accelerator - brake set. ORBEX was used, which is an experimental fuzzy coprocessor. The reason for using this is that it permits to write fuzzy rules as sentences in natural language. Also, ORBEX allows the user to define any variables.

Goerick et al. (1996) presented a feature based car detection and system. The authors have taken with reference to the CARTRACK system (Brauckmann, Goerick, Gross, & Zielke, 1994). The feature-based car detection is used to detect, measure, and also to track front and rear view of automobiles. For this to be possible, images should be captured from the viewing point of the following car in the form of a sequence. The author calls it a 'monocular visual sensor system'. The two parts of the detection system as discussed by the authors are pre-processing and classification/detection modules. Integral treatment of image regions and the pre-processing method determine the speed of the approach. The classification and detection task is done by Artificial Neural Networks (ANNs). ANNs are used because they give fast results and hence right decisions can be taken by the machine. Thus using ANNs are more efficient.

According to the research conducted by Goerick et al. (1996), five laser range finders were used to measure the distances of objects. Time series based attention control is one of the various applications discussed by the authors to determine the centre of attention, which is the most important beam for Automated Cruise Control (ACC). This was based on the velocity of the test vehicle used and distance information. The research concluded that time series were heavily disturbed by noise. For solving this, the problem was classified into two parts. The solutions were achieved by a modular neural network. According to the authors for every beam, a confidence measure of the current distance value is determined. This is based on the current and following distance values. The confidence values vary in the range of zero to one, meaning it ranges from an unreliable distance to reliable distance information. But, Kalman filtering (Gelb, 1974) opposed, where a non-model-based approach is followed. The confidence measures are computed by single hidden layer neural networks implementing the mapping. The neural networks are trained to deal with invalid, unknown data and noise.

The first part of the processing is the non-linearity of the network, which allows for a very spontaneous computation of valid confidence values. The next part of the processing is the fusion or the combination of the confidential information, the distance information, and the velocity information to determine the focus of attention. These two processing stages are characterised by the type of data being processed, in its respective stages, which are temporal information and spatial information (of the first and second stage). Hence neural networks simplify the implementation of reducing or nullifying the complexity.

Saffiotti (1997) said that to bring a robot to a target position, the only effective way is to track a pre-computed path. This is when the following two components are verified: All the assumptions that are used during the computation of the path/way are valid even during execution, like certain parameters of the environment (soil condition, wind conditions, etc.) has not changed; the autonomous vehicle is able to establish a relationship between itself and the path that it has to follow. These conditions are usually not met in the real world applications, hence many scientists and researchers prefer sensor-based behaviours. A sensor-based behaviour implements a control policy based on external sensing, and spontaneously the vehicle moves with respect to certain features in the environment, instead of moving with respect to a represented internal path. This type of vehicular/robotic control using sensor-based behaviour is also called as compliant control. A few examples are moving along a wall, avoiding obstacles, etc. Another example which was recorded earlier is dated back to 1985 when Sugeno and Nishida (1985) developed a fuzzy controller which was able to drive a car along a path between two walls. Here, a rotating SONAR sensor was used to measure the distances from the two walls. An 8080 microprocessor was used to run the fuzzy controller. Not only was this successful, but also there was a disadvantage, that is the controller was not robust with respect to the errors in the sensor. It was also reported that the speed was slow (1.7 cm/s). Also, the tolerance to sensor errors did not see any improvement. After this paper was published, a fuzzy controller to avoid obstacle was developed by Takeuchi.

The authors used simple algorithms to obtain data about unoccupied areas and occupied areas ahead of the robot from a video camera. The experiment was successful, but there were a few errors in the visual system, which led to the failure of the robot at times. Later many authors incorporated fuzzy sensor based behaviours in their vehicles. Murphy and Hawkins (1996) proposed a different approach called 'tactical behaviour'. A tactical behaviour acts as a supervisory controller, observing and collecting data from the recent execution of the machine in order to modify some navigation parameters. Later another behaviour called 'memory processing' was proposed by Pin and Bender (1997). The memory processing behaviour's task is to manage an internal state used to track, detect, and escape from limit cycles.

A few autonomous robots, namely MORIA, FLAKEY, MARGE, and LOBOT included these fuzzy behaviours for performing various tasks like crossing a door, reaching a particular location, etc. Hence all the behaviours can be applied to autonomous vehicles for better functioning.

Fuzzy control systems are thus credited, for designing robust controllers that deliver good performance when there are disturbing parameters like noise. Autonomous vehicles or robots need these characteristics where:

• An exact mathematical model or detail information of the environment is not available.
• The data from the sensor is imprecise and also uncertain.
• Real-time operation is required.

Sugeno and Nishida's (1985) car model is an example of fuzzy control. All these are possible because in fuzzy control it is easier to write simple behaviour which could be used to do multiple tasks. It does not require any kind of complex mathematical models. The qualitative behaviour or nature of fuzzy control is that it can be transferred and moved to different platforms, without a or less modification to the system. The smooth and easy movement of the robot is because of its interpolative nature, where the vehicle can move between two specified points easily. Deep learning can also be used to enhance the performance of fuzzy control systems.

Recently, Saquib, Ashraf, and Malik (2017) discussed the methods to detect traffic signs using Cascaded Neural Network (CNN). CNN starts to learn with only one input node, and it proceeds by adding new inputs, this is followed by adding new and hidden neurons too. The traffic signs detection is based on Multi-Layer Perception Neural Networks. According to the authors' research, there are two neural network forms, that is blue and red neural network forms. The neural networks were trained using real images of traffic signs. According to their research, the neural networks were trained using Back-Propagation with Momentum algorithm, which is an iterative method for optimizing a different objective function. Each traffic sign represents or corresponds to a neural network. The authors conducted an experiment where the networks were learned in this software. In the near future, as technology improves, the vehicle becomes similar to a computer. The authors said that the vehicles will become a computer with tyres and though there may be traffic congestion or vehicle blocking, there will be no serious accidents or fatalities. Lately, new complex assistance systems like 'advanced driver assistant systems' are incorporated to enhance the autonomous vehicle's performance. Certain sensing techniques are also of high importance.

Artificial intelligence, especially neural networks, machine learning, and deep learning have become an absolute necessity to make autonomous vehicles function properly and safely. All these make the vehicles more efficient and does not pose a threat to neither the pedestrians nor the travellers, with advanced sensors and technology autonomous vehicles can predict what might happen and hence gather information to perform the necessary tasks. Autonomous vehicles also reduce distracted driving accidents to a great extent.