Vehicular automation
Vehicular automation is using technology to assist or replace the operator of a vehicle such as a car, truck, aircraft, rocket, military vehicle, or boat. Assisted vehicles are semi-autonomous, whereas vehicles that can travel without a human operator are autonomous. The degree of autonomy may be subject to various constraints such as conditions. Autonomy is enabled by advanced driver-assistance systems of varying capacity.
Related technology includes advanced software, maps, vehicle changes, and outside vehicle support. The benefits of viewing automated driving from a sociotechnical systems perspective has been discussed.
Autonomy presents varying issues for road, air, and marine travel. Roads present the most significant complexity given the unpredictability of the driving environment, including diverse road designs, driving conditions, traffic, obstacles, and geographical/cultural differences.
Autonomy implies that the vehicle is responsible for all perception, monitoring, and control functions.
SAE autonomy levels
Technology
Software
Autonomous vehicle software generally contains several different modules that work together to enable self-driving capabilities. The perception module ingests and processes data from various sensors, such cameras, LIDAR, RADAR, and ultrasonic SONAR, to create a comprehensive understanding of the vehicle's surroundings. The localization module uses 3D point cloud data, GPS, IMU, and mapping information to determine the vehicle's precise position, including its orientation, velocity, and angular rate. The planning module takes inputs from both perception and localization to compute actions to take, such as velocity and steering angle outputs. These modules are typically supported by machine learning algorithms, particularly deep neural networks, which enable the vehicle to detect objects, interpret traffic patterns, and make real-time decisions. Furthermore, modern autonomous driving systems increasingly employ sensor fusion techniques that combine data from multiple sensors to improve accuracy and reliability in different environmental conditions.Perception
The perception system is responsible for observing the environment. It must identify everything that could affect the trip, including other vehicles, pedestrians, cyclists, their movements, road conditions, obstacles, and other issues. Various makers use cameras, radar, lidar, sonar, and microphones that can collaboratively minimize errors.Navigation
Navigation systems are a necessary element in autonomous vehicles. The Global Positioning System is used for navigation by air, water, and land vehicles, particularly for off-road navigation.For road vehicles, two approaches are prominent. One is to use maps that hold data about lanes and intersections, relying on the vehicle's perception system to fill in the details. The other is to use highly detailed maps that reduce the scope of real-time decision-making but require significant maintenance as the environment evolves. Some systems crowdsource their map updates, using the vehicles themselves to update the map to reflect changes such as construction or traffic used by the entire vehicle fleet.
Another potential source of information is the environment itself. Traffic data may be supplied by roadside monitoring systems and used to route vehicles to best use a limited road system. Additionally, modern GNSS enhancement technologies, such as real-time kinematic and precise point positioning, enhance the accuracy of vehicle positioning to sub-meter level precision, which is crucial for autonomous navigation and decision-making.
History
The "Stanford Cart", created by Hans Moravec in the late 1970s while he was a graduate student, was the first experimental autonomous vehicle. It was a precursor to both NASA's Moon and Mars Lander projects as it was known at the time that radio signal lag times would have made anything other than autonomous control impractical. The box rested on 4 bicycle wheels, and had a camera, battery and a radio antenna connecting it wirelessly to a remote computer. It could also be steered remotely. Morovic was able to get the Cart to navigate around large obstacles in a 100 foot long room, albeit it would take 5 hours as the cart would frequently stop as the computer processed images which it analyzed and then responded with navigation instructions.Approximately 20 years later the Robotics Lab at Carnegie Mellon University developed ALVINN a vehicle with 3 onboard Sun Microsystems computers that, using a camera and a laser range finder, could slowly drive itself down a road by monitoring the white divider line.
Automated vehicles in European Union legislation refer specifically to road vehicles. For those vehicles, a specific difference is legally defined between advanced driver-assistance system and autonomous/automated vehicles, based on liability differences.
AAA Foundation for Traffic Safety tested two automatic emergency braking systems: some designed to prevent crashes and others that aim to make a crash less severe. The test looked at popular models like the 2016 Volvo XC90, Subaru Legacy, Lincoln MKX, Honda Civic, and Volkswagen Passat. Researchers tested how well each system stopped when approaching moving and nonmoving targets. It found that systems capable of preventing crashes reduced vehicle speeds by twice that of the systems designed to mitigate crash severity. When the two test vehicles traveled within 30 mph of each other, even those designed to lessen crash severity avoided crashes 60 percent of the time.
Sartre
The SAfe Road TRains for the Environment project's goal was to enable platooning, in which a line of cars and trucks follow a human-driven vehicle. Trains were predicted to provide comfort and allow the following vehicles to travel safely to a destination. Human drivers encountering a train could join and delegate driving to the human driver.Tests
Self-driving Uber vehicles were tested in Pittsburgh, Pennsylvania. The tests were paused after an autonomous car killed a woman in Arizona. Automated busses have been tested in California. In San Diego, California, an automated bus test used magnetic markers. The longitudinal control of automated truck platoons used millimeter wave radio and radar. Waymo and Tesla have conducted tests. Tesla FSD allows drivers to enter a destination and let the car take over.Risks and liabilities
Ford offers Blue Cruise, technology that allows geofenced cars to drive autonomously.Drivers are directed to stay attentive, and safety warnings are implemented to alert the driver when corrective action is needed. Tesla, Incorporated has one recorded incident that resulted in a fatality involving the automated driving system in the Tesla Model S. The accident report reveals the accident was a result of the driver being inattentive and the autopilot system not recognizing the obstruction ahead. Tesla has also had multiple instances where the vehicle crashed into a garage door. According to the book "The Driver in the Driverless Car: How Your Technology Choices Create the Future," Tesla automatically performs an update overnight. The morning after the update, the driver used his app to "summon" his car, and it crashed into his garage door.
Another flaw with automated driving systems is that unpredictable events, such as weather or the driving behavior of others, may cause fatal accidents due to sensors that monitor the surroundings of the vehicle not being able to provide corrective action.
To overcome some of the challenges for automated driving systems, novel methodologies based on virtual testing, traffic flow simulation and digital prototypes have been proposed, especially when novel algorithms based on Artificial Intelligence approaches are employed which require extensive training and validation data sets.
Implementing automated driving systems poses the possibility of changing built environments in urban areas, such as expanding the suburban regions due to the increased ease of mobility.
Challenges
Around 2015, several self-driving car companies including Nissan and Toyota promised self-driving cars by 2020. However, the predictions turned out to be far too optimistic.There are still many obstacles in developing fully autonomous Level 5 vehicles, which is the ability to operate in any conditions. Currently, companies are focused on Level 4 automation, which is able to operate under certain environmental circumstances.
There is still debate about what an autonomous vehicle should look like. For example, whether to incorporate lidar to autonomous driving systems is still being argued. Some researchers have come up with algorithms using camera-only data that achieve the performance that rival those of lidar. On the other hand, camera-only data sometimes draw inaccurate bounding boxes, and thus lead to poor predictions. This is due to the nature of superficial information that stereo cameras provide, whereas incorporating lidar gives autonomous vehicles precise distance to each point on the vehicle.
Technical challenges
- Software Integration: Because of the large number of sensors and safety processes required by autonomous vehicles, software integration remains a challenging task. A robust autonomous vehicle should ensure that the integration of hardware and software can recover from component failures.
- Prediction and trust among autonomous vehicles: Fully autonomous cars should be able to anticipate the actions of other cars like humans do. Human drivers are great at predicting other drivers' behaviors, even with a small amount of data such as eye contact or hand gestures. In the first place, the cars should agree on traffic rules, whose turn it is to drive in an intersection, and so on. This scales into a larger issue when there exists both human-operated cars and self-driving cars due to more uncertainties. A robust autonomous vehicle is expected to improve on understanding the environment better to address this issue.
- Scaling up: The coverage of autonomous vehicles testing could not be accurate enough. In cases where heavy traffic and obstruction exist, it requires faster response time or better tracking algorithms from the autonomous vehicles. In cases where unseen objects are encountered, it is important that the algorithms are able to track these objects and avoid collisions.