We developed a model predictive control -based simulator using differential equations and realizing a final simulation loop system mimicking vehicle dynamics, driver behavior, road geometry, and a decision-making system. The simulator is developed in Wolfram Mathematica and relies on a symbolic numerical approach to optimize a vehicles track (represented as a cost function) for autonomous driving enabling lane-keeping, collision-avoidance, path-planning, and decision-making features while assessing and reducing the risk associated to the vehicle by providing warnings and interventions when a crash is imminent.
1. Safe Driving Advisor and Evaluator
(Vehicle Risk Assessment and
Control for Lane-Keeping and
Collision Avoidance)
PRESNTED BY: HAZEM MOHAMED FAHMY
SUPERVISED BY: DR. MOHAMED ABD-ELGHANY
CO- SUPERVISED BY: PROF. GERD BAUMANN
2. We usually get to learn from our mistakes, but
not when driving - the road is an exception!
2
Mate, I'm so sorry. I thought there was time.
You just pulled out, I don't have time to stop.
Oh c'mon mate. It was a simple mistake.
I know, if I was going a bit slower, but...
Please, I've got my boy in the back.
I'm going too fast.
I'm sorry.
3. Outline
Introduction
Background
o Passive Safety Systems
o Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
3
4. Outline
Introduction
Background
o Passive Safety Systems
o Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
4
5. Introduction
5
Lane-
keeping
A study by
Kuehn et
al., 2014 [3]
Germany
Lane
departure
caused
14% of
accidents
Caused
30% of
fatalities
Automotive
Safety
European
Road Safety
Observatory
Report,
2008 [1]
European
Union
1.8 million
injuries
43,000
fatalities
Collision
Avoidance
NHTSA
Report,
2013 [2]
CA is the
most type
of crashes
USA Police
Report,
2013
CA causes
29% of
accidents
6. Outline
Introduction
Background
o Passive Safety Systems
o Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
6
7. Background
7
Automotive Safety
Research Field
Safety Systems
Passive
Safety
Post-accident
(Injury reduction)
Active Safety
Pre-accident
(Prediction)
Implementation
Techniques
Risk
Assessment
Real-time
Algorithm
Sensor
Fusion
Raw data
fusion
8. Outline
Introduction
Background
Passive Safety Systems
o Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
8
9. Passive Safety Systems – Internal Airbags
9
Illustration of the process of deploying an Airbag. [4]
Aim
Reduce impact force of
occupant caused by inertia
followed by an accident.
Type
Injury
Reduction
Implementation
A soft air bag that
absorbs and endure the
impact force of an
occupant.
10. Passive Safety Systems
– Pedestrian Airbags
10
Pedestrian Airbag deployment made by Volvo. [5]
Aim
Reduce impact force of a
pedestrian when hit by a
car
Type
Injury
Reduction
Implementation Sensors to get
information about
the pedestrian
Airbags that are installed
at the front of the vehicle
11. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
11
12. Active Safety Systems – Dynamic Active
Display (DAD)
12
Dynamic Active Display used icons [6]
Aim
Minimize deviation of
driver’s gaze directions
Type
Visual-
Advisory
Implementation
Icons presented using
a special windshield
for laser applications
Icons classified into Warning,
Numbers, Graphical
13. Active Safety Systems – Adaptive Cruise
Control (ACC)
13
Adaptive Cruise Control illustration [7]
Aim
Maintain safe distance
with the preceding vehicle
Type
Intervention
Implementation
Fuzzy logic algorithm
Sensor Fusion technique
Throttle and Braking valve
controller
14. Active Safety Systems – Emergency Lane
Assist (ELA)
14
Road departure scenario where “H” denotes
the host vehicle. [8]
Aim
Prevent risky lane
departure
Type
Intervention
Implementation
Risk Assessment system
that estimates the risk
based on Sensor Fusion
Activation system that decides
and intervenes based on the
risk level previously estimated
15. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
o Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
15
16. Description and Implementation - Critique
The previous mentioned work for Passive Safety systems only took action
after an accident to reduce injury.
Some Active Safety systems were only able to give the driver an advisory
signal.
Other Active Safety systems were designed for lane-keeping or collision
avoidance only.
The proposed system presents the idea of combining collision avoidance
and lane-keeping features into one system. The idea is proved using
mathematical modeling and mathematical-based risk assessment
algorithm.
16
17. Description and Implementation – Cont’d
17
Block Diagram of the proposed system
Driver Model
(Normal/Rough
Schemes)
Vehicle Model
(Set of 2nd order
differential
equations)
Road Model (Lane-
keeping/Collision
Avoidance)
Vehicle motion
and parameters
monitoring
Optimization and
Control Algorithm
Set of constraints
and parameters
Validated
Vehicle
Model
18. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
o Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
18
19. Vehicle Model – Equations and Diagram
19
Newton’s equations of motion.
The goal is to be able to control and
monitor the motion of a vehicle – as
close to reality as possible.
The used model is a four-wheel
vehicle model which is assumed by
previous related work to have very
reliable behavior. Modeling notation of the vehicle forces
in longitudinal and lateral directions.
Also shows the rotational and
transitional velocities.
22. Vehicle Model – Assumptions
22
Assumption 1:
Rear steering angles are assumed to be zero. Front steering angles are
assumed to be equal.
23. Vehicle Model – Assumptions
23
Assumption 2:
Longitudinal velocity ẋ is assumed to never settle at steady-state.
24. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
Driver Model
o Road Model
o Optimization and Control
Results
Summary and Conclusion
24
25. Driver Model
Previously discussed driver model in P. Falcone et al. [9] was used during this
work. The generation of driving schemes depends on gains, and orientation
error of the vehicle to the road.
Image processing of gains and orientation error was done using Matlab.
The image processed data was then fed to Mathematica to output the result.
The output result was then fed to the vehicle model in order to validate it.
25
26. Driver Model – Driving Schemes
26
Input gains and
steering wheel output
of a normal driving
scheme
Steering wheel output
after computing the
image-processed data
27. Driver Model – Driving Schemes
27
Input gains and
steering wheel output
of a rough driving
scheme
Steering wheel output
after computing the
image-processed data
28. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
Driver Model
Road Model
o Optimization and Control
Results
Summary and Conclusion
28
29. Road Model – Lane-Keeping
The environment of the vehicle was
modeled as a polynomial function
with k and m randomly assigned to
output an appropriate road
environment.
29
30. Road Model – Collision Avoidance
In Collision Avoidance, interpolation is
done using Hermite technique to generate
an obstacle.
Interpolation is done from x0 to xte with a
width Є. The three parameters could be
altered to output any kind of obstacles.
The following constraint was used in
order to assure an appropriate obstacle
behavior and collision avoidance scenario.
30
31. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
Driver Model
Road Model
Optimization and Control
Results
Summary and Conclusion
31
37. 37
Optimization and Control – Cont’d
Comparison between cost function with and
without implementing penalty function
38. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
Driver Model
Road Model
Optimization and Control
Results
Summary and Conclusion
38
39. Results – Catching and Lane Keeping at mid-speed
39
Vehicle constants used to
describe the vehicle used
Initial conditions and parameters
used during optimization process
40. Results – Catching and Lane Keeping at mid-speed
40
Vehicle motion with respect to
road geometry
Cost function change during
optimization process
41. Results – Catching and Lane Keeping at mid-speed
41
Steering angle change during the
optimization process (-3.14,3.14)
Braking ratio change during
optimization process (1,-1)
42. Results – Catching and Lane Keeping at mid-speed
42
Deviation quantity change
during optimization process
Orientation quantity (alignment
error) change during optimization
43. Results – Catching and Lane Keeping at mid-speed
43
Steering angle and Braking ratio discrete points at every
discretization time samples (an optimization parameter)
44. Results – Catching and Lane Keeping at mid-speed
44
Optimization constraints limits
specified during the process
Scenario evaluation result for
constraints violation
45. Results – Collision Avoidance at low-speed
45
Vehicle constants used to
describe the vehicle used
Initial conditions and parameters
used during optimization process
46. Results – Collision Avoidance at low-speed
46
Vehicle motion with respect to
road geometry
Cost function change during
optimization process
47. 47
Results – Collision Avoidance at low-speed
Steering angle change during the
optimization process (-3.14,3.14)
Braking ratio change during
optimization process (1,-1)
48. 48
Results – Collision Avoidance at low-speed
Deviation quantity change
during optimization process
Orientation quantity (alignment
error) change during optimization
49. 49
Results – Collision Avoidance at low-speed
Optimization constraints limits
specified during the process
Scenario evaluation result for
constraints violation
50. Outline
Introduction
Background
Passive Safety Systems
Active Safety Systems
Description and Implementation
Vehicle Model
Driver Model
Road Model
Optimization and Control
Results
Summary and Conclusion
50
51. Summary and Conclusion
Vehicle model, Road model, Driver model, Optimization and Control Algorithm
were investigated for the work setup
The results shows collision avoidance of smooth and rough obstacles.
Also shows lane-keeping of straight, parabolic, and inclined (negatively, and
positively) lanes at low velocity.
For high velocities, the vehicle was controlled to stay in an inclined lane and
off-track situation.
51
52. Future Work
The future work should concentrate on the computation speed to reach in-the-
loop-simulation level and furthermore, on hardware level – as having an
optimization parameter that describes the computational speed.
Another optimization parameter that could be added is the rate of change of
steering angle that will allow a more smooth process of control.
Different tire models should be investigated to ease the process of optimization
and insure stability.
The influence of the obstacle structure should also be investigated and studied
on the optimization performance.
52
53. Comparison to Previous Work
The closest approach to our proposed system was found to be in A. Gray et al.
in 2013 under the title “A Unified Approach to Threat Assessment and Control
for Automotive Active Safety”, IEEE Transactions on Intelligent Transportation
Systems, Vol. 14, September 2013.
This work used Model Predictive Control as the optimization algorithm control
concept of the research which is not reliable if compared to the used algorithm
in our proposed system. If the algorithm failed to converge to a solution it will
malfunction instead of switching to another technique.
This work’s main interest was lane-keeping which does not guarantee 100%
vehicle safety as an accident will occur if an obstacle is to be presented in the
pre-defined road geometry. The proposed system presented lane-keeping and
collision avoidance combined into one system.
53
54. Comparison to Previous Work
Furthermore the mentioned work only intervenes when the risk of departing a
lane is estimated to be high. The proposed system takes control of the vehicle
during the whole optimization process.
54
The optimization algorithm intervenes
when the vehicle is close to eymax or eymin
The vehicle is controlled to be strictly
overlapping the center of lane
55. References
1. E. R. S. Observatory, “Annual statistical report,” SafetyNet, 2008.
2. N. T. Report, “Analysis of light vehicle crashes and precrash scenarios based
on the 2000 general estimates system,” Springfield, 2013.
3. M. Kuehn, T. Hummel, and J. Bende, “Analysis of car accidents caused by
unintentional run off road,” German Insurers Accident Research, 2014.
4. Mercedes Benz, 1 ed., 2014.
5. L. Jakobsson, T. Broberg, and H. Karlsson, PEDESTRIAN AIRBAG TECHNOLOGY
A PRODUCTION SYSTEM. Volvo Car Corporation, 1 ed., 2015.
55
56. References
6. A. Doshi, S. Y. Cheng, and M. Trivedi, “A novel active heads-up display for
driver assistance,” IEEE Transactions on Systems, Man, and Cybernetics, Part
B (Cybernetics), vol. 39, no. 1, pp. 85–93, 2009.
7. P. Worrawut, T. Somphong, and P. Manukid, “Adaptive cruise control for an
intelligent vehicle,” 06 2014.
8. A. Eidehall, J. Pohl, and F. Gustafsson, “A new approach to lane guidance
systems,” Proceedings. 2005 IEEE Intelligent Transportation Systems, 2005.,
2005.
9. P. Falcone, M. Ali, and J. Sjoberg, “Predictive threat assessment via
reachability analysis and set invariance theory,” IEEE Transactions on
Intelligent Transportation Systems, vol. 12, no. 4, pp. 1352–1361, 2011.
56
57. Acknowledgment
This work was performed on the computational resource bwUniCluster funded
by the Ministry of Science, Research and the Arts Baden-Wrttemberg and the
Universities of the State of Baden-Wrttemberg, Germany, within the framework
program bwHPC.
I would also like to thank Dr. Hassan Mostafa and Eng. Ali for allowing me using
their image processing code of Matlab which was useful in extracting the real-
time data of different driving schemes.
57