Autonomous driving, more and more consumers are paying attention to and accepting self-driving cars
Automobile automatic driving has been a hot spot in the automobile industry in recent years. Automobile automatic driving technology refers to the use of computers, sensors, controllers, and other high technology to realize the technology of unmanned vehicles.
With the help of self-driving technology, vehicles can autonomously sense their surroundings, automatically plan their paths and avoid obstacles to achieve a driverless effect. Some high-end car brands such as Tesla and Mercedes-Benz have already launched their self-driving technology, and more and more consumers are beginning to pay attention to and accept self-driving cars.
Tesla is an industry leader in the practice of Autopilot technology. Tesla Autopilot is divided into Tesla AP (autopilot) and FSD (Full Self-Drive).
Autopilot Automated Assisted Driving allows the vehicle to automatically assist with steering, automatically assist with acceleration, and automatically assist with braking in the lane. The Fully Self-Driving Capability Suite (FSD) includes additional features and can be continuously enhanced through software updates.
In November 2018, Tesla owners have driven more than 1 billion miles (about 1.6 billion kilometers) with Autopilot enabled, and by 2022 that figure will have surpassed 3 billion miles (more than 4.8 billion kilometers), which is the equivalent of circling the globe 120,000 times. Those 3 billion miles are from actual roads, not computer simulations.
Tesla with Autopilot averages one accident per 9.17 million kilometers throughout 2022, which is 773% safer than human driving. If you compare the Tesla FSD when it’s on to the national average accident rate, you can conclude that Tesla Full Autopilot is about 4.94 times safer than a human driver. So what Tesla officials are trying to show is that the safety numbers for both Autopilot and Full Self-Driving FSD far exceed the national average.
With the continuous development of technology and the in-depth application of advanced control algorithms, the future development of automobile automatic driving will show the following trends.
Trend 1: Automobile automatic driving technology will develop in the direction of level 4 and 5 automatic driving. This level of automatic driving will be completely free of human intervention, realizing the real driverless.
Trend 2: Automotive self-driving technology will gradually promote from high-end cars to low-end cars, reduce the cost of popularization, and the degree of popularization will continue to expand.
Trend 3: Automotive self-driving technology will gradually expand to more fields, such as public transportation, logistics and distribution, which will lay a solid foundation for the realization of urban intelligent transportation systems.
Automatic driving technology has made remarkable progress after years of research and development. At present, automatic driving technology has realized the leap from primary to advanced, including different levels such as L1 (partially automated), L2 (partially automated + monitored), L3 (conditionally automated), L4 (highly automated) and L5 (fully automated). Among them, L4 and L5 are the core of current autonomous driving technology and the future development direction.
Autonomous Driving Levels: The International Society of Automotive Engineers (SAE) standard is divided into six levels, L0~L5; the U.S. National Highway Traffic Safety Administration (NHTSA) divides autonomous driving into five levels.
A framework for automated driving technology:
(1) Estimation of vehicle sideslip angle
Vehicle state information plays a key role in motion planning, decision-making, and control techniques for AVs and CAVs. State includes vehicle position, velocity (longitudinal and lateral velocity in vehicle coordinates), and attitude (vehicle pitch, roll, and yaw). Accurate knowledge of longitudinal and lateral velocities to calculate the vehicle sideslip angle is critical to vehicle control. While expensive equipment, such as the RT3000 and Kistler S- S-S-Motion, can measure some of these state variables, they are not feasible for mass-produced vehicles. Therefore, many scholars have been working on estimating the vehicle sideslip angle in real automotive applications.
(2) Trajectory tracking control of self-driving vehicles
Vehicle Trajectory Tracking Control is a key and fundamental component of AVs, calculating direct optimization commands such as steering angle, throttle opening and brake pedal to replace and assist the driver. The control module ensures that the vehicle follows a predetermined path accurately while avoiding obstacles and maintaining safe driving conditions. Therefore, micro-level feedback or self-learning tracking control algorithms are designed to adjust the vehicle’s speed, acceleration, and steering to follow the desired trajectory. Typically, the algorithms consider the dynamics and kinematics of the vehicle to determine the optimal control inputs. These inputs are adjusted in real-time to account for any uncertainties or disturbances in the vehicle environment, such as changes in road conditions, traffic flow, or the presence of obstacles.
(3) Co-control of CAVS
Equipped with sensors such as cameras, radar and LIDAR are key components of intelligent transportation systems. However, despite the advanced sensor technology available in AVs, it is not always possible to fully and reliably sense dynamic and variable environments using only onboard sensors, due to adverse weather conditions, uncertainty in sensors or vehicle models, and changes in lighting.
The current status of automated driving can be divided into two aspects: technology and market. In terms of technology, the current automatic driving technology has been able to realize level 2 and 3 automatic driving, for example, Tesla’s automatic assisted driving system (ADAS) can automatically control basic operations such as acceleration, braking and steering. As for the market, automated driving technology has been recognized by consumers to a certain extent.
Automatic driving technology has a wide range of application fields, including public transportation, cabs, logistics transportation, personal travel and so on. In the field of public transportation, self-driving buses and cabs have been gradually put into operation; in the field of logistics transportation, self-driving trucks can improve transportation efficiency and safety; in the field of personal travel, self-driving cars can improve travel efficiency and comfort.
Tesla’s FSD represents industry-leading technology. FSD is a full-link autopilot hardware and software architecture that includes sensing/control/execution.
FSD Architecture: We have built a set of hardware and software architecture of the whole chain of automated driving, including perception, regulation and control, and execution at various levels of data, algorithms, and computing power. Planning: The essence is to solve the multi-object association path planning problem, handle the trajectories of self and all objects, and guide the car to complete the corresponding execution actions. Neural Networks (Neural Networks): analyzes video streams and other information, and outputs complete kinematic states (position/speed/acceleration/bumps) to control the vehicle. Training Data: Forms a closed loop of data through the latest 4D auto-labeling technology, upgraded simulation and cloud computing resources. Training Infra: including CPU, GPU, Neural Network Accelerator, AI compiler, etc., of which the AI compiler can support the new operations required by neural networks and map them to the best underlying hardware resources. Compiler & Inference: How to run neural networks on computers. Current inference engines can distribute the execution of a single neural network across two separate systems on a chip, which can be thought of as two separate computers connected within the same self-driving computer.
Tesla FSD V12 (FSD V12.1.2) has begun its official push to users. “FSD Beta V12 upgrades the urban street driving stack to an end-to-end neural network, trained on millions of videos, replacing over 300,000 lines of C++ code.” So says Tesla in the update notes. In other words, the upgrade to FSD V12 will see AI replace engineer coding and take charge of vehicle behavior in Autopilot mode.
In June 2019, Tesla officially pushed the latest version of NOA to all models equipped with FSD Full Self-Driving in the Chinese market, which enables the vehicle to automatically drive into and out of highway on-ramps or overpass off-ramps and overtake slow-moving vehicles. In June 2023, Tesla’s founder Elon Musk announced that Tesla’s Autopilot technology, FSD (Full Self-Driving), would be open for use by other automakers. Since opening up their patents for free a few years ago, Tesla has been trying to help other automakers. Now, they are sharing the Supercharger network and are more than willing to license Autopilot or FSD to other automakers. This move will undoubtedly help speed up the entire automotive industry so that more automakers can benefit from Autopilot technology. Whether it will be an industry-shaking force like when Tesla opened up all its patents for electric cars, 2024 is waiting to see.
