Autonomous cars are very closely linked to Industrial IoT. IoT, combined with other technologies such as machine learning, artificial intelligence, local computing, etc., provides the essential technology for autonomous cars. For many the very inquisitive question is how these autonomous cars work. What is actually working inside to make them work without the driver taking control of the wheel. It is well known that these days cars are equipped with a lot of sensors, actuators and controllers. These end devices are driven by software that sits on different function-specific software running on ECUs ( Electronic Control Units). Machine learning software is also part of the set.
One of the main tasks of any machine learning algorithm in a self-driving car is the continuous rendering of the surrounding environment and the prediction of possible changes in the surroundings. These tasks are divided mainly into four sub-tasks: Detection of objects Object Detection or Recognition Classification of objects Item Localization and Movement Prediction
Machine learning algorithms can be divided into four categories: regression algorithms, pattern recognition, cluster algorithms, and decision matrix algorithms.
Algorithms of Regression
In ADAS, images (radar or camera) play a very important role in localization and operation, while the biggest challenge for any algorithm is to create an image-based model for prediction and feature selection. The type of regression algorithms that can be used for self-driving cars include Bayesian regression, neural network regression, and forest decision regression, among others.
In ADAS, images obtained from sensors have all kinds of environmental data; filtering of images is needed to identify instances of an object category by removing irrelevant data points. Pattern recognition algorithms are effective at removing uncommon data points. Recognizing trends in a data set is an important step before classifying objects. These algorithm types can also be described as data reduction algorithms.
These algorithms help to reduce the data set by detecting object edges and modifying line segments (polylines) and circular arcs to the edges.
Line segments are aligned to the edges of the corner, so a new line segment is started. Circular arcs are ideal for line segment sequences that approximate an arc. The features of the image (line segments and circular arcs) are combined in a number of ways to shape the features that are used to identify the object.
Often the images generated by the device are not clear and it is difficult to detect and locate objects. It is also possible that the classification algorithms can miss the object and fail to identify it and report it to the system. The explanation may be low resolution images, very few data points or discontinuous data. This type of algorithm is good at discovering the data point structure. Like regression, it defines the problem class and the process class. Clustering techniques are usually structured by modelling approaches such as centroid-based and hierarchical. Both approaches are concerned with the use of inherent data structures to better organise data into classes with highest commonality. The most widely used algorithm type is K-means, a multi-class neural network.
Matrix Algorithms of Decision
This type of algorithm is good at systematically defining, evaluating and ranking the output of the relationship between values and knowledge sets. These algorithms are primarily used for decision-making. If a car wants to turn left or to brake depends on the degree of faith the algorithms have in the classification, identification and prediction of the next movement of objects. These algorithms are models made up of several decision models that are independently trained and whose predictions are combined in some way to make the overall prediction while at the same time reducing the probability of decision making errors. The most widely used algorithms are GDM and AdaBoosting.
While technologies and capabilities continue to evolve to make autonomous vehicles a reality, there are some hurdles. Right now, self-propelled cars are legal only in a few U.S. states, as regulators measure how best to ensure their safe interaction with standard human-driven vehicles.
“Self-driving capability will add benefits to our entire society, such as providing transportation for people who are otherwise unable to drive due to age or physical impairment,” says Freckmann. “That’s both exciting and meaningful.”