SoC chip selection for intelligent driving domain controller-EEWORLD

Collect

The intelligent driving industry chain consists of the perception layer, decision layer, and execution layer. The vehicle-mounted perception system of the perception layer mainly includes cameras, ultrasonic radars, lidars, millimeter-wave radars, etc.; the roadside assistance system mainly includes high-precision maps, satellite positioning, inertial navigation, and V2X technology. The decision layer mainly includes ADAS algorithms, vehicle-mounted chips, vehicle-mounted storage, high-precision maps, and cloud platforms. The execution layer mainly includes electronic drive, electronic steering, electronic braking, and lighting. The platform layer mainly includes big data, intelligent driving solutions, traditional vehicle networking, and smart cockpits. The terminal composition mainly includes vehicle-mounted OBU, road test unit RSU, mobile phone APP, and edge computing.

With the development of the intelligent driving industry, intelligent driving functions are becoming increasingly complex, and the penetration rate of mainstream intelligent driving assistance functions has steadily increased. Lane keeping assistance, emergency braking assistance, adaptive cruise, blind spot monitoring, lane centering assistance, lane merging assistance, automatic lane change assistance, automatic parking, memory parking and other functions have gradually been implemented. The intelligent driving system has an increasingly strong demand for sensors and computing power. The intelligent driving system requires a large amount of computing power and multiple types of computing resources. The typical intelligent driving system processing flow is as follows.

An intelligent driving system is a vehicle control system that can autonomously perceive, make decisions, and execute driving tasks. The levels of intelligent driving are usually based on the standards of the International Organization for Standardization (ISO), and are divided into six levels from L0 to L5. These levels reflect the maturity and degree of automation of the autonomous driving system. The following is a detailed description of each level:

L0: No automation. The driver is fully responsible for the driving process, and no automated functions are involved. This is the traditional driving method and does not involve autonomous driving technology.

L1: Assisted driving function. The vehicle provides some automated functions, such as adaptive cruise control and automatic parking, but the driver still needs to bear the main driving tasks and responsibilities. Adaptive cruise control is an intelligent cruise system that can automatically adjust the vehicle speed and driving posture according to the road conditions ahead, reducing the driver's driving pressure. Automatic parking is an automatic parking assistance system that can automatically complete the parking process after the vehicle finds a suitable parking space, without the driver having to operate the steering wheel, accelerator and brake. In addition, the lane keeping function is also an important part of L1 level autonomous driving. It can identify road boundaries through cameras and sensors, and automatically adjust the vehicle's driving trajectory to keep the vehicle stable in the lane. These functions can assist the driver in completing some driving tasks and improve driving convenience and safety.

L2: Partial Automation. The vehicle can autonomously complete certain driving tasks under specific conditions, mainly including adaptive cruise control, automatic parking, lane keeping, lane change assist and automatic lane change functions. These functions can reduce the driver's driving pressure to a certain extent and improve the convenience and safety of driving. The driver still needs to monitor the driving environment and be ready to take over. Among them, adaptive cruise control and automatic parking functions have already appeared in L1 level autonomous driving, and have been further improved and perfected in L2 level. In addition, the lane keeping function is also more intelligent and mature in L2 level autonomous driving. It can automatically identify road boundaries and adjust the vehicle's driving trajectory to keep the vehicle stable in the lane. Lane change assist and automatic lane change functions can help the driver automatically complete lane change operations at the right time to improve driving safety.

L3: Conditional Automation. Under certain circumstances, the vehicle can drive autonomously most of the time. The driver does not have to pay attention all the time, but needs to take over in time when the system requests it. This level of autonomous driving can complete the identification and driving of the vehicle's surrounding environment under certain circumstances, and make autonomous decisions and perform corresponding operations based on the collected data. In addition to conventional functions, it also includes more advanced autonomous driving functions such as automatic parking functions for more complex scenarios, automatic driving on highways, and memory navigation. Specifically, it can realize the vehicle's automatic lane recognition, adaptive cruise, automatic lane change and other functions in specific scenarios, and even automatically complete overtaking actions on highways. In addition, it can also autonomously plan travel routes based on navigation information, and complete autonomous overtaking and other actions under certain circumstances.

L4: Highly automated. The vehicle can drive autonomously in various environments and conditions. The driver does not need to participate in driving most of the time, and the system can autonomously handle complex road conditions under specific circumstances. In some cases, the driver may not need to do anything. The main intelligent driving functions include adaptive cruise control, automatic parking, traffic congestion assistance, lane departure warning, and vehicle active obstacle avoidance. When the vehicle is in a complex environment or uncertain situation, it can automatically identify lanes and handle emergencies, such as automatic lane change and automatic avoidance. In addition, it can also autonomously plan travel routes based on navigation information, and autonomously complete driving tasks under specific circumstances, such as automatically entering highways and other scenarios. In general, the L4 level autonomous driving system already has a high degree of intelligence and autonomous decision-making capabilities.

L5: Fully automated. This is the highest level of autonomous driving, where the vehicle can drive completely autonomously in any environment and conditions, and the driver can be completely freed from the driving task inside or outside the vehicle. The vehicle can drive and operate autonomously like a robot. Fully autonomous driving: The L5 level autonomous driving system can drive completely autonomously on various roads and environments without any human intervention. Autonomous route planning: The system can autonomously plan the optimal route according to the navigation target, and automatically select the road, adjust the speed, avoid obstacles, etc. Intelligent perception and decision-making: Through a variety of sensors such as high-definition maps, lidar, and cameras, the system can fully perceive the surrounding environment and make autonomous decisions and judgments. Complex scene processing: The system can handle driving tasks in complex scenes, including highways, urban roads, rainy and snowy weather, and has the ability to respond to emergencies. Safety assurance function: The system has a variety of safety assurance measures, such as automatic braking, obstacle avoidance, lane keeping, etc., to ensure the safety of passengers and pedestrians. At present, many companies and research institutions are working hard to develop related technologies towards this goal. It should be noted that despite the rapid development of technology, there are still many technical and regulatory issues to be solved to achieve L5 level autonomous driving.

The entire intelligent driving system processing process usually involves the following types of computing resources:

Deep learning: Environmental perception modules are the largest users of deep learning computing power, including various common image and laser point cloud detection algorithms, such as object detection, lane line detection, traffic light recognition, etc., which involve a large number of typical neural network (NN) operations. Such modules are usually implemented using highly customized NN accelerators.
Visual processing: This type of module is computationally intensive but not deep learning algorithmic modules, such as image signal processing (ISP), image pyramid, distortion correction (Rectify), local feature extraction, optical flow tracking, image encoding and decoding (Codec), etc. This type of module usually uses hardened dedicated visual accelerators to achieve low latency.
General computing: Although customized deep learning and visual processing accelerators can meet most common mature computing-intensive operations, they still cannot cover all needs. With the rapid development of cutting-edge technologies and the deepening of self-developed technologies, a considerable number of customized computing modules will often be generated. Such modules are usually also computationally intensive operations that cannot be efficiently implemented using CPUs, so general computationally intensive processing units (such as DSPs and GPUs) are also required to implement them.
Logical operations: This type of module contains a large number of logical operations and is not suitable for implementation using computationally intensive processors. It is generally implemented using general-purpose CPU processors. This type of module includes common multi-sensor perception fusion algorithms (such as Kalman filter KF), optimization-based decision-making and planning algorithms, vehicle control algorithms, system-level functional logic, diagnostic logic, shadow mode data mining functions, etc.