? SELF DRIVING CARS
Self-driving cars also known as robot cars, autonomous cars, Driverless cars. A Car which is capable of sensing the environment and moves with little or no input of human being.
These cars consist of many sensors that provide the exact navigation paths like Sonar, GPS, odometer, RADAR, Lidar. Advance control system. This technology is capable of providing convenient and safe driving by avoiding crashes caused by human errors. It is critical to ensure that autonomous vehicles interact with human-driven vehicles safely. Driving uniformity is an important factor for safety. It is a significant challenge for autonomous cars to imitate, while staying within safety bounds, human driving styles, i.e., to achieve human-like driving. Human driving provides passengers with comfortable driving, and confidence that the car can drive independently, enable surrounding drivers to better understand and predict autonomous vehicles behavior so that they can interact with it naturally. To achieve human-like driving, it is useful to introduce a driver model that reproduces individual drivers behaviors. Replicating driving trajectories is one of the primary objectives of modeling longitudinal motion in traffic flow. Known as car-following models, these models are essential components of microscopic traffic simulation.
First, limited accuracy. Most current car-following models are simplified, i.e., they contain only a small number of parameters. Simplification leads to suitable analytical properties and rapid simulations. However, it also renders the models limited in flexibility and accuracy because using few parameters can hardly model the inherently complex car-following process.
Second, poor generalization capability. Car-following models calibrated with empirical data try to emulate drivers output by finding the model parameters that maximize the likelihood of the actions taken in the calibration dataset. As such, the calibrated car-following model cannot be generalized to traffic scenarios and drivers that were not represented in the calibration dataset.
Third, absence of adaptive updating. Most parameters of car-following models are fixed to reflect the average drivers characteristics. If car-following models with such pre-set parameters are used for autonomous vehicles, they can reflect neither the driving style of the actual drivers of the vehicles nor the contexts in which they drive.
To address the above limitations, a model is proposed for the planning of human-like autonomous car following that applies deep reinforcement learning. Deep RL, which combines reinforcement learning algorithms with deep neural networks to create agents that can act intelligently in complex situations. Deep RL has the promise to address traditional car-following models limitations in that (1) deep neural networks are adept at general purpose function approximates that can achieve higher accuracy in approximating the complicated relationship between stimulus and reaction during car-following; (2) reinforcement learning can achieve better generalization capability because the agent learns decision-making mechanisms from training data rather than parameter estimation through fitting the data and (3) by continuously learning from historical driving data, deep RL can enable a car to move appropriately in accordance with the behavioral features of its regular drivers.Why do we need autonomous cars?
Safety is one of the important concerns for human being as when human drivers are driving cars, they can get detracted by phone calls, songs, games and other distraction like sometimes a person can be in a bad mood or just had a fight that effect the driving concentration and increases the chances of accident. As these cars can communicate with each other that also reduces the chances or ends the chances of accidents. EQUITY is another major consideration. Self-driving technology could help mobilize those individuals who are unable to drive themselves, such as the elderly, those who dont know how to drive or disabled. But the widespread adoption of autonomous vehicles could also displace thousands of Indians employed as drivers, negatively impact public transportation funding, and perpetuate the current transportation systems injustices. ENVIRONMENTAL impacts are a serious concern these days, and a major uncertainty. Accessible and convenient self-driving cars could increase the total number of miles driven each year. If those vehicles are powered by gasoline or any other hazardous substances then transportation-related climate emissions could skyrocket. If, however, we use low-power chips for the cars then we can find a perfect solution for environment. To the extent that electrified self-driving cars enable more shared rides, emissions could drop even further. SAVING TIME is also important as self-diving cars will save a lot of time as ride will be exact and car will tell us the exact time and will reach on time as computer is driving not a human so it can give a uniform speed. COMFORT as when elders or kids travel in cars due to bad roads, they can feel jerks or discomfort these cars will give the comfort and can access the upcoming jerks as a human vision is not very far computer can detect. We could eliminate the traffic signals then there will be less stops results in comfort and saves time too. As these cars dont need a driver so in highly populated areas where we face parking issues this car can drop the passenger and can circle around till when the passenger is ready to leave