1. S245408 Waiyuntian Lou
S253778 Huanyu Su
Under the instruction of:
Prof. Massimiliana Carello
Lane Keeping Assist (LKA)
Academic Year 2018-2019
14/12/2018
2. Chapter 1 Introduction
2
-LKA is a feature that take step to keep the vehicle stay in its
lane. If it detect that the vehicle is drifting out of the lane, it may
gently steer the vehicle back into it.
What is Lane Keeping Asist system (LKA/LKAS)
- According to NHTSA, about 80% of the accidents are caused by the
drivers.
- 20% of traffic accidents are caused by departure of vehicle from its lanes.
- Up to 26% of all relevant accidents with injuries and fatalities can be
prevented by lane keeping support.
Why?
3. Chapter 1 Introduction
3
- The LKA system is a auxiliary function, not a autonomous
function. The driver should always touch the steering wheel.
Drivers with hands-off is a misuse.
- The activation condition of LKA system is at least 60km/h. The
curvature should be at least 250m.
UN/ECE Regulations
[1] Regulation No 79 of the Economic Commission for Europe of the United Nations (UN/ECE)
— Uniform provisions concerning the approval of vehicles with regard to steering equipment
[2] Regulation No 130 of the Economic Commission for Europe of the United Nations (UN/ECE)
— Uniform provisions concerning the approval of motor vehicles with regard to the Lane Departure Warning System (LDWS).
4. Reference
4
[1] Guo Hongqiang, Chen Hui, Chen Jiayu. Design of Lane-based Lane Maintenance System Based on EPS[J].
Automotive Technology, 2018(08): 33-38.
[2]Yu Lijiao. Design and experimental verification of lane-based auxiliary control algorithm based on EPS [D]. Jilin
University, 2016.
Structure of LKA control system
[1]Baharom M B, Hussain K, Day A J. Design of full electric power steering with enhanced performance over that of
hydraulic power-assisted steering[J]. Proceedings of the Institution of Mechanical Engineers Part D Journal of
Automobile Engineering, 2013, 227(3):390-399.
[2] Zhang Hailin. Lane keeping system based on electric steering [D]. Tsinghua University, 2012.
[3]Cheng Shuliang. Modeling and simulation analysis of electric power steering system [D]. Chongqing University,
2016
EPS model
[1]Prof. Nicola Amati. Chassis A notes. High speed cornering simplified approach
[2]Yu Lijiao. Design and experimental verification of lane-based auxiliary control algorithm based on EPS [D]. Jilin
University, 2016.
2 d.o.f. vehicle model
5. Reference
5
[1] Carlo Novara. Automatic control. Lecture 21 Lane Keeping
[2] Zhang Hailin. Lane keeping system based on electric steering [D]. Tsinghua University, 2012.
Driver model
[1] Ziegler, J.G & Nichols, N. B. (1942). "Optimum settings for automatic controllers’’ Transactions of the ASME. 64:
759–768.
[2] Nyquist, H. (1932). "Regeneration Theory". Bell System Tech. J. USA: American Tel. & Tel. 11 (1): 126–147
Design of PID controller
7. Chapter 2 System structure
7
- It is the information acquisition system, includes various
sensors and image processing modules
Sensing layer
- Including the information processing, the lane departure
warning algorithm, the driver operation state identification
algorithm and the lane keeping active control algorithm.
Decision layer
- Uses the steering system or the braking system to control the
vehicle motion
Execution layer
8. Chapter 2 Sensing layer
8
- Steering wheel angle sensor
- Steering torque sensor
Steering
- Acceleration sensors
- Yaw rate sensor
- Wheel speed sensors
Dynamic of the vehicle
- Camera is commonly used.
Lane information & car’s relative position
9. Chapter 2 Execution layer
9
- Audible
- Visual
- Tactile
Execution for warning
- Electric Power Steering (EPS).
- Electric Stability Program (ESP).
Execution for steering
12. Chapter 3 State Decision Strategy
12
- Shutdown
- Standby
- Intervention
Target: define system state
- α----Activation condition coefficient
- β----Intervention condition coefficient
In order to identify the system state, we introduce 2
auxiliary coefficients.
13. Chapter 3 State Decision Strategy
13
- Determine when the LKA system should
intervene the vehicle control.
- Reduce false warning.
- Avoid collisions between LKA system and the
driver.
This strategy is important
14. Chapter 3 Criteria for activation
14
- Switch state (ON/OFF)
- The clarity of the lane marking.
- Lane change (Ex. Turning light is on)
3 criteria should be taken into consideration
The target is to define the value of coefficient α
15. Chapter 3 Criteria of intervention
15
The target is to define β
- For intervention:
𝑇𝐿𝐶 < 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
- The system control the vehicle until the
vehicle steer back to lane centre.
The criteria for intervention and for exit are
different.
TLC -----a time interval estimated according to the Time to Line Crossing
algorithm. Which indicate that if the car maintains the current dynamic
condition, after TLC seconds the left/right front wheel will touch the lane
marking.
18. Chapter 3 Path error controller
18
Target road centerline function Ytarget =0
Predicted driver model
θtarget depends on predicted lateral displacement(the
predicted lateral displacement of vehicle a certain time)
21. Chapter 3 EPS model
21
States:
Current of electric motor Ie
Angle of motor θe
Rotation speed of motor
Inputs:
Voltage of electric motor U
Equivalent self aligning torque Tsa
State space equation of EPS model
WHAT
Why.-----It is mostly for safety reasons. It is critical especially for sleepy or distracted drivers. Here is some statistics.
80%, ……
20%.........
So, nowadays, some countries have already introduced regulations to promote this feature.
For deeper studying, we decided to reconstructed a LKA system by using Simulink and Carsim.
Then, we read some papers. The structure of the control system is based on these two papers.
Following shows the models we used and the papers where these models come from.
we used a 2 d.o.f. vehicle model, EPS model, driver model and PID controller.
Decision layer Execution layer we will introduce later.
In order to realize the lane keeping feature. Sensors are required to obtain lane information, vehicle dynamic and steering information.
Execution layer
2 types of execution ,first is the execution for warning.
When the system detect that the vehicle has a risk of running out of the lane, the system will alert the driver first.
It can alert the driver by using audible devices, for example alarms, or visual devices, for example, a screen in dashboard.
But, the most effective way is by using tactile warning. For example the vibration of steering wheel or seat.
Then, if the driver does not react in time. The LKA will control the vehicle and steer it back to the lane center. This action is accomplished by using these two types of executions. EPS or ESP, or we can using them together for steering.
This is the structure of the decision layer. As you can see, two controllers inside it which are State decision controller and Path error PID controller. Now, I will focus on the state decision controller. Why we need state decision block? Because the LKA system shouldn’t always turned on. For example it should swithced off when the driver intends to change the lane. So we need this block to verify the interfere condition.
The target of this strategy is to define the system state. The system has 3 states. (Shutdown, standby and intervention)
念ppt
The picture on your right hand side represent the logic of the strategy.
If the activation condition is not satisfied, alpha=0,the system is shutdown. Otherwise, alpha=1.
When alpha=1, the system is activated. Then, if the intervention condition is not satisfied, beta=0, the system standby. When beta=1, the system shift to intervention state and start controlling the vehicle.
This strategy is important.
Why?
Firstly, through state decision, this strategy determine when the system should intervene the vehicle control and when the system should just standby and keep monitoring the vehicle status.
In this way, we can reduce false warning and avoid collision.
Now, what we need to do is to define the value of alpha and beta.
The flowchart represent the logic for the estimation of alpha.
As you can see, 3 criteria should be considered. The first is the switch state. It means that there is a interface on the dashboard which the driver can turn on or turn off the system.
The second criterion is the clarity of the lane marking. As I said before, the system is highly depend on the lane marking. If the marking is not clear, the system should not be activated.
The third criterion is lane change. It means that, if the driver want to change lane, the system should not interfere the operation.
The problem is how to judge whether the driver is intentionally change lane or not.
Then , How to define beta.
So, TLC is a time that after which the vehicle will drift out of the lane.
Now let’s have an overview of the whole control loop, my colleague has already introduced the state decision block. Once it decides that LKA system should interfere, the path error PID controller will request a target steering angle from the EPS module which is proportional to the lateral displacement.
After that the electric power steering module which contains a EPS controller and EPS model(which we will discuss it later) will determine the actual steering angle base on the required steering angle. We notice that the self-aligning torque is also an input for the EPS module because we need it to solve the dynamic equation of the steering system.
Finally the actual steering angle will be the only input to the Vehicle model, here we exploit the simplest 2 d.o.f model which doesn’t consider the self-aligning torque and the aerodynamics. The output which is the vehicle dynamic information will be the input of the state decision block.
Now let’s have an overview of the whole control loop, my colleague has already introduced the state decision block. Once it decides that LKA system should interfere, the path error PID controller will request a target steering angle from the EPS module which is proportional to the lateral displacement.
After that the electric power steering module which contains a EPS controller and EPS model(which we will discuss it later) will determine the actual steering angle base on the required steering angle. We notice that the self-aligning torque is also an input for the EPS module because we need it to solve the dynamic equation of the steering system.
Finally the actual steering angle will be the only input to the Vehicle model, here we exploit the simplest 2 d.o.f model which doesn’t consider the self-aligning torque and the aerodynamics. The output which is the vehicle dynamic information will be the input of the state decision block.
First the assumption is that the road centerline function in the global reference frame is known. In this case we consider the vehicle is on a highway so the road is straight. The road centerline function is simply Y=0.While the actual vehicle path function in the global reference frame can be determined from the vehicle dynamic information.
Here we exploited a predicted driver model. Because during this project we found that if we determine the required steering angle based on the current lateral displacement, the control is not stable, because the steering correction is too late. So according to the predicted driver.The required steering wheel angle is not base on the current lateral displacement, but based on the predicted lateral displacement after for example 0.5 second. And the stability problem is solved.
Now let’s have an overview of the whole control loop, my colleague has already introduced the state decision block. Once it decides that LKA system should interfere, the path error PID controller will request a target steering angle from the EPS module which is proportional to the lateral displacement.
After that the electric power steering module which contains a EPS controller and EPS model(which we will discuss it later) will determine the actual steering angle base on the required steering angle. We notice that the self-aligning torque is also an input for the EPS module because we need it to solve the dynamic equation of the steering system.
Finally the actual steering angle will be the only input to the Vehicle model, here we exploit the simplest 2 d.o.f model which doesn’t consider the self-aligning torque and the aerodynamics. The output which is the vehicle dynamic information will be the input of the state decision block.
EPS system contains EPS model and EPS PID controller.
The from this electric circuit equation. Here we assume that the steering mechanism is rigid, and we neglect the damping, friction of the steering system. We can write the dynamic equation of the electric motor and in this equation we need the self-aligning torque equivalent at the electric motor shaft to solve this equation.
After that we are able to write the state space equation of the EPS model. From the this equation we are able to derive the actual electric motor angle and thus the actual steering angle. The states are….. And the inputs are…
About the PID controller of the EPS, we also use a simple Proportional control. The input voltage of the electric motor is proportional to the steering angle error. A problem is how to define Kp, of course for a PID control you can tune the parameter to have a trade-off between overshoot and response time. But first you should know Kp is in which order of magnitude, it’s around 0.1 or 100? In this case we solve the transfer function of this system and determine Kp so that this system is in critical damping, after we tune Kp around this value However, in this case we found that optimum Kp for EPS is not the optimum Kp for the whole system. In order to have a more stable behavior of the vehicle, we should have a short response time of EPS.
Now let’s have an overview of the whole control loop, my colleague has already introduced the state decision block. Once it decides that LKA system should interfere, the path error PID controller will request a target steering angle from the EPS module which is proportional to the lateral displacement.
After that the electric power steering module which contains a EPS controller and EPS model(which we will discuss it later) will determine the actual steering angle base on the required steering angle. We notice that the self-aligning torque is also an input for the EPS module because we need it to solve the dynamic equation of the steering system.
Finally the actual steering angle will be the only input to the Vehicle model, here we exploit the simplest 2 d.o.f model which doesn’t consider the self-aligning torque and the aerodynamics. The output which is the vehicle dynamic information will be the input of the state decision block.
Here we only exploited the simplest 2 d.o.f car model not considering self-aligning torque and aerodynamics which is not precise. Later we will see that there are some differences between the results when we substitute the 2 d.o.f model with a more precise model imported from CarSim.
The structure of the simulink model is just the same as that we have introduce before. T
The test condition is 60km/h and between first and second seconds we apply a disturbance to the steering system.
Blue line is the lateral displacement with respect. Red curve is yaw angle psi, orange curve is the steering wheel angle in rad. And purple line indicates the LKA interfere condition, at high level means LKA is interfering. Here we can see that after the disturbance is applied to the steering wheel at 1s, LKA doesn’t interfere immediately, since the vehicle is still near the centerline. When the vehicle continuosly deviates from the centerline LKA starts to interfere and we can see a obvious correction of the steering wheel angle in the orange line. Which is due to the activation of the electric power steering. And also, we can see with the correction the lateral displacement and yaw angle decreases to zero which means the vehicle is back to the centerline.
After designing the control algorithm based on the 2 d.o.f vehicle model. It’s possible to have a verification by substituting the vehicle model with a more precise model imported from CarSim. We just replace the vehicle model with CarSim model, we also need to set the input and output of this model to compatible with the algorithm.
This curve indicates the lateral displacement when we use 2 d.o.f and CarSim as the car model. The curve with CarSim has a smaller oscillation and longer response time. Since we only use a simple unprecise 2 d.o.f, the difference is quite large. Maybe after if we have time we can try to exploit 2 d.o.f model consideration self-aligning torque and aerodynamics or even 10 d.o.f model to see if the difference can be reduced.
Finally I want to make a conclusion with what we have learned in this project.
Studied the LKA and SAE levels
Studied and Realized the state decision strategy using Stateflow
Designed the path error controller
Established EPS model and designed the EPS controller
Established vehicle model
Verified the algorithm
解释一下stateflow图形函数就是logic flow