Autonomous vehicles are typically classified into different levels of automation, from L0 to L5:
L0: No automation, the vehicle is completely controlled by the driver, and there are no automatic driving assistance functions.
L1: Driver assistance, the vehicle has some basic assistance functions, such as cruise control and lane departure warning.
L2: Partial automation, where the vehicle can perform lateral and longitudinal control simultaneously, such as adaptive cruise control and lane keeping assist.
L3: Conditional automation, where the vehicle can drive itself under specific conditions, but the driver must be ready to take over at any time.
L4: High level of automation, where the vehicle drives itself fully within a limited area and under specific conditions, without driver intervention.
L5: Fully automated, the vehicle can drive itself in any scenario, even without a driver's seat.
The core technologies of autonomous vehicles include sensors and artificial intelligence. Sensors such as LiDAR and cameras are used to perceive the surrounding environment and provide information such as road conditions and obstacles; artificial intelligence is used for decision-making and control to ensure the safe and efficient operation of the vehicle.34
Inertial sensors, accelerometers, and gyroscopes, acting as the "inner ear" of a car, have been performing basic, low-performance tasks for years in applications such as airbags and stability control systems.
Without the input of other sensors, inertial sensors can independently detect the motion of a vehicle. For example, a simple single-axis accelerometer can detect a car's rapid deceleration at high g-forces to deploy airbags. More advanced inertial sensor assemblies consist of two orthogonal XY accelerometers and a single-axis/dual-axis angular velocity gyroscope, commonly used for vehicle stability control. Lateral and longitudinal accelerations and rotational speeds determine whether the vehicle needs to take measures to prevent rollover or reduce wheel slippage during cornering. Essentially, inertial sensor assemblies determine whether a car can travel on a predetermined path set by the driver.
While typical automotive inertial sensor assemblies measure motion in a single direction, an inertial measurement unit (IMU), an embedded module of three-axis linear accelerometers and three-axis angular velocity gyroscopes, measures six degrees of freedom ("6 DOF" or six-axis). By combining linear motion (three-dimensional space) and rotational measurement components (roll, pitch, and yaw) into a six-axis structure, the IMU captures all components of the vehicle's motion. Beyond airbags and vehicle stability control, the IMU can track and calculate the vehicle's position and orientation in real time. Therefore, after precise calibration to eliminate temperature and bias drift, combined with an extended Kalman filter algorithm, the IMU can accurately locate the vehicle in a short time without any additional assistance. More advanced systems integrate wheel speed and angle information to aid Kalman filter positioning estimation, further improving positioning accuracy.
The latest generation of Advanced Driver Assistance Systems (ADAS) and autonomous vehicles require high-precision IMUs to predict vehicle motion and determine real-time position. In these advanced systems, IMU information is fused with GPS receivers and even visual sensors (such as LiDAR and cameras) to continuously estimate and update vehicle position information, which is then input into the system's central computing module. This type of IMU navigation system, which relies on fusing additional sensor data (such as GPS), is called an Inertial Navigation System (INS).
A GPS receiver alone cannot provide continuous, high-precision location information. However, by receiving satellite signals distributed globally, it can pinpoint location information to within a few meters. After correcting for satellite clock errors and atmospheric propagation errors, a GPS receiver can use algorithms such as Real-Time Kinematics (RTK) to achieve a positioning accuracy of approximately 2–4 centimeters. GPS receivers typically update location information once per 1 Hz or per second, but can also achieve 10 Hz to 20 Hz to meet the needs of dynamic positioning applications. In short, under optimal road conditions, a vehicle on a highway will update its GPS location information approximately every 10 feet.
An IMU (Integrated Measurement Unit) can be used to estimate a vehicle's position between each update from a GNSS/GPS receiver, thereby increasing the frequency of positioning information output. Furthermore, GPS receivers often lose signal in environments "unfavorable to GPS signals" (such as near tunnels and tall buildings). In these cases, the IMU needs to estimate the position for 10, 20, or even 30 seconds. The longer the estimation time, the greater the error in the IMU's position estimation. Typical autonomous driving systems allow for a positioning error range of only 10–30 cm. While some military and research IMUs can currently provide this performance, their exorbitant prices (up to five figures) make them prohibitively expensive.
To offer affordable IMUs to the market, developers typically use MEMS-based accelerometers and gyroscopes. Mass-produced silicon-based MEMS IMU sensors are priced below $100, better meeting the cost requirements of consumer and industrial systems. Next-generation MEMS IMU sensors promise to provide the accuracy and reliability required for advanced automotive applications, including fully automated Level 5 drive systems.
MEMS-based inertial measurement units (IMUs) offer sizes and manufacturing processes suitable for the automotive market. Several high-performance MEMS IMU sensors currently on the market have gyroscope bias instability (BI) of 5°/h, angular random walk (ARW) of 0.5°/√h, and acceleration BI in the 10µg range. These products can provide effective and smooth position information between GPS update intervals. However, when vehicles pass through tunnels or underground passages, these mid-performance IMUs struggle to maintain position accuracy below 10 cm after a few seconds. The most advanced MEMS inertial sensors are currently striving to achieve gyroscope BI close to 1°/h and ARW of 0.1°/√h. Once this technological level is reached, GPS+IMU combined navigation systems will be able to meet the performance requirements of high-level autonomous driving applications.
In addition to the issues mentioned above, vibrations or high instantaneous accelerations can cause the silicon microfinger of the accelerometer and gyroscope structure in MEMS sensors to stick together. Due to van der Waals forces, once stuck together, they are very difficult to separate. The device cannot resolve this issue through power-on/power-off cycles like other semiconductor devices.
While the consumer and industrial markets can accept the failure rate of MEMS-based gyroscopes/accelerometers, can they meet the automotive industry's demands for low failure rates and long service life? This will be another major challenge for IMU device developers.
Everyone eagerly anticipates the day when self-driving cars will replace existing older vehicles, freeing up valuable space in garages and parking lots and making our roads safer and more efficient. Currently, tens of thousands of engineers worldwide are working on developing next-generation sensing technologies to achieve this milestone in the transportation sector as soon as possible.
IMU device technology and INS navigation technology may not be familiar to the general public, but they are undoubtedly fundamental elements to ensure the safety and efficiency of autonomous vehicles.
Autonomous driving is an intelligent vehicle system that relies on advanced technologies and aims to achieve driverless operation. It is also known as driverless cars or wheeled mobile robots.
Autonomous driving relies on the collaborative efforts of artificial intelligence, computer vision, radar technology, monitoring devices, and the Global Positioning System (GPS) to ensure vehicles can operate safely without human intervention. Devices such as video cameras, radar sensors, and laser rangefinders are used to perceive the surrounding traffic environment and combine this with detailed map data for navigation.
Google's data centers process the vast amounts of information about the surrounding terrain collected by these vehicles, making them appear as remote-controlled or intelligent cars controlled by Google's data centers. This technology relies on the support of various advanced technologies, such as artificial intelligence and computer vision. Through the combination of these technologies, autonomous vehicles can intelligently adapt to various road conditions, improving driving safety and efficiency. Autonomous driving technology is a key application of the Internet of Things (IoT). Prior to this, there was no standard for classifying autonomous driving levels in China. This standard was mainly based on the US SAE standard, and they are basically the same, with some adjustments made to suit the Chinese context. There are three main points: First, the SAE standard places non-driving automation functions and safety assistance functions in Level 0, referred to as "no automation," while in China it is called "emergency assistance." This primarily allows the driver to take control of the vehicle, while the system can perceive the environment and provide assistance, warnings, and even brief intervention, treating it as a fundamental safety feature. Second, the Chinese standard stipulates that Levels 0-2 autonomous driving require the driver and system to collaborate in response to events, not that all actions are performed by the driver as in the SAE standard. Third, the Chinese standard explicitly adds the risk of driver takeover in Level 3 autonomous driving, including takeover capability monitoring and risk mitigation, and clarifies minimum safety requirements to reduce practical application risks.
