Design of steer-by-wire software architecture for lateral control requirements of autonomous driving

1. Lateral control requirements for autonomous driving
The lateral control requirements for autonomous driving are somewhat different from those of traditional actuators. In the actual process, such as steering or braking, many domestic R&D are currently reverse developed or developed based on the current state, lacking a forward development concept. All actuators and wire control components are designed to meet the needs of drivers or vehicle motion control. Therefore, when designing steering, there must be a forward thinking method, not to edit the MAC verification method, but to design wire control components based on the needs of drivers and autonomous driving. Therefore, we must first know what the chassis needs to do for autonomous driving, or what kind of operation the driver needs the chassis to do. At present, the development of steering is basically divided into two levels. One is the electric power steering EPS less than or equal to L2, but in addition to the original auxiliary power system, some wire control interface instructions are added to EPS. There is also a wire control steering greater than or equal to L3, which has its particularity at this stage. If it is less than or equal to L2, the main function safety will be turned off when it fails, which is fail off. For the state greater than or equal to L3, it is necessary to fail-operationl, that is, to ensure certain functions or degrade functions when it fails, and basically have steering functions, which is a very important critical point.

Software Defined Chassis

In order to realize the above functions, various technologies have been proposed. First of all, we need to have a positive mindset. Autonomous driving has certain requirements for the chassis. The software-defined chassis on the left is proposed by our research institute and Qingche Zhixing. The definition of the software chassis needs to be based on the needs of autonomous driving or the needs of an independent mobile chassis. In the future, the development of the whole vehicle may be a chassis plus a cockpit, that is, three domains and two platforms. The three domains support autonomous driving, the intelligent cockpit domain, and the wire-controlled chassis domain; the two platforms are two hardware platforms, the intelligent cockpit and the intelligent chassis. With the continuous changes in the shape of the intelligent cockpit, we must first define various cockpit forms, that is, the stability of various cockpits is an indicator to be defined. Only under this indicator, we can consider the planning instructions for autonomous driving and define the current state of the whole vehicle, whether it is a stable state, an unstable state, or a critical stable state, that is, a safety level prediction model. After the model is defined, it can maximize the guarantee that horizontal and vertical coordination is actually implemented. After the horizontal and vertical coordination instructions come out, there is also software definition to define the hardware and software decoupling of the execution system. Because the chassis domain is a large computing platform and algorithm center, the underlying steering and braking, and even the suspension and drive-related application layer algorithms can be continuously iterated on the platform, so the execution system wire control and system hardware and software decoupling are required, and then the horizontal and vertical coordination related instructions are defined. At the same time, the software-defined chassis can also define the extended functions of autonomous driving. This is the demand for software-customized chassis. Our research institute has realized the relevant architectures of safety level prediction, dynamic control, and hardware and software decoupling based on chassis domain control greater than or equal to L3. The wire control execution system on the right has different requirements for meeting chassis control requirements and vehicle requirements, especially for steering redundancy design and functional safety, including high-level road feel experience requirements, and later there are independent steering-related function requirements. In terms of technical route, it is from software-defined chassis to domain, and then to execution components. In the actual production process, the product must be implemented, and it must be reversed based on the execution components to meet the technical requirements of each level. This is the software-defined chassis.

Autonomous driving classification

As autonomous driving and driver assistance systems continue to mature, the classification of autonomous driving has undergone several rounds of optimization. The latest version of SAE driving automation levels pays close attention to how L3 is achieved. Last year, when SAE clearly proposed L3, a person must be driving when a function is requested. If the vehicle does not make a request, if a safety accident or other danger occurs to the vehicle, the responsibility lies with the vehicle. Once a function request is issued, someone must take over directly. Different companies have different views on how to achieve the time delay between them. In terms of safety risk prediction, when a function is requested, there must be a certain buffer time for the driver to perform related operations, because in theory, L3 can take both hands off the steering wheel, take the feet off the brake pedal, and even read the phone/newspaper for a short time. When the vehicle has safety risks that it cannot overcome on its own, a request is made to the driver, and the driver must take over. In the process of continuous practice, the buffer time needs to be continuously optimized and updated, mainly for safety or risk prediction time, or advance response time, so L3 is also a very tangled stage. Some people even think that it should be 2.99 or skip L3 directly, but L3 is still an inevitable process in the short term, so many studies based on steering are between L3 and L2 or L3. The direct switch between automatic driving and manual driving in L3 is a big test, and the Switch Ramp processing method for switching between automatic driving and manual driving is very important. When it comes to L4 and L5, there are also some typical states. L4 still has a steering wheel, but it is a completely decoupled state, and the L5 steering wheel no longer exists. For steering, the mainstream R&D goals and R&D paths of each company are also different. There are two types in Europe. One is that when it comes to L4, the steering wheel can be cancelled and replaced with a driving terminal, that is, an operating handle, for emergency switching. Some still retain the steering wheel, and the design ideas are different, so the Switch Ramp problem between L3 and L3 to L2 must also be considered at the current stage.

Horizontal control development trend

What is horizontal development? If we consider the chassis control level of autonomous driving, there may be three-axis direction control in the future, namely, horizontal, longitudinal and vertical control. Steering is the most important operation in horizontal control, including automatic steering, so horizontal control is an important control level for later development. The automotive industry is developing towards electrification, intelligence and networking. Software-defined cars have been promoted. The domain control electronic and electrical architecture has developed rapidly from distributed to domain control, then to centralized domain, and even to the central computer mode in the later stage. As a key lateral control mechanism, the steering system is the direction of future development. There will be several requirements for the future transfer system. One is the road feeling simulation and performance requirements. The L3 to L4 stage is a very important stage. At this time, there is also a steering wheel, and it is a completely decoupled steering wheel. At this time, there will be some requirements, such as weak vibration noise reduction, variable steering ratio, and motion control, which need to be reflected in the road feeling simulation and performance requirements. The requirements for the control-by-wire chassis include flexible arrangement of steering and steering wheel, combined design based on chassis domain, and cross-platform application. From the perspective of safety and reliability, there are fail-operation, redundant architecture and functional safety. In terms of autonomous driving, the transition between L2 and L3 involves angle control and steering wheel silencing/folding.

Steering requirements for autonomous driving levels

The demand for steering in autonomous driving is completely based on the current level of autonomous driving, and requirements are put forward for steering characteristics. First, from the conventional level 0 to level 1 and level 2, the driver and the driver system are the main focus. Conventional electric processing steering uses a simple three-phase motor, independent power supply, position sensor and one or two core controllers, which may not have heterogeneous redundancy, because below L2 is in fail off state, and it will be turned off once danger or failure occurs. Dual-core functions are required in terms of functional safety level, especially when it cannot be turned off, the ASIL D safety level must be achieved, that is, it must be turned off when it must be turned off. Quadrant protection is also to meet ASIL D. When there is a phase short circuit or motor short circuit, it can be turned off in time to ensure the relevant communication, network security, and mechanical coupling status. There will be a big change to L3. The person responsible for driving is the system, and the entire EPS design, that is, the wire control steering design, is a redundant EPS actuator design. The steering system has crossed over to wire control steering, which has two parts: one is road feel simulation, and the other is steering brake control or steering execution. It requires a six-phase motor or dual-motor configuration, dual power supply, dual TAS backup, and three cores. Of course, it can also be two functional cores, plus a lock-step core to ensure the functional safety level and safe implementation. When the safety level reaches ASIL D, there must be heterogeneous redundancy measures, including expected functional safety factor analysis, quadrant protection, multiple communication modes, and hardware safety module modules to ensure related communications. At this time, mechanical coupling is cancelled, but there is still a steering wheel. At L4 and L5, the steering wheel can be shielded, and even some control terminals are used for implementation. There are two voices in Europe, one is the steering wheel, and the other is the terminal, but the American and Japanese and Korean systems still prefer to use the steering wheel and do not want to use the terminal method.