Neural Network Based Lane Change Trajectory ... - SUNY Plattsburgh

muscleblouseΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 4 χρόνια και 2 μήνες)

103 εμφανίσεις

Presented by Rob Schulz


CSC485


Spring 2012


SUNY Plattsburgh ~ rschu003@gmail.com

1)
Terms / Keywords

2)
Problems

3)
Solutions

4)
Results

5)
Issues

6)
Conclusion

7)
References

Sections

Artificial Neural Network (ANN
)



Prediction system which uses input to
determine output by constant updates and reconfigurations. Often called
“feed
-
forward”


Autonomous Vehicle



Vehicles which use external data to drive itself with
no human intervention


Intelligent Transportation
Systems



the infrastructure, vehicles, users, and
devices networked together to provide better/more advanced driving
abilities and prevention of accidents

Keywords

Vehicle
Trajectory



Where a vehicle is planning on
going

Driver
Behavior



How a driver drives based on their experiences
& training

VANETs



Vehicle Ad
-
Hoc
Networks

Traffic Simulation &
Communication Network
Simulation



simulations on
how the network and traffic will act/react

Performance
Evaluation



Evaluation how well networks and devices work

Dynamic
Route
Planning



Constantly updated traffic routes

Keywords

1)
Planning

2)
Preparation

3)
Crossover

4)
Adjustment

Lane Change Process

What is not a lane change

What is a lane change

Lane Changes

Yi Yang,
Rajive

Bagrodia

1)
Different trainings

2)
Human errors

3)
A
lane change process may cause disturbance
in the
traffic

4)
Costs are high



Problems

-
Collision Avoidance Systems

-
Artificial Neural Networks

-
Intelligent Transport Systems

-
Cooperative Systems

Solutions

1)
Predictions

2)
Gathering of information

Collision Avoidance
Systems

1)
Information
obtained is
limited

A.
Blinded reality

I.
Inclusion of vehicles

II.
Sensor capabilities

Collision Avoidance Systems
-

Issues

Modeling the possible outcomes of lane changes
and their possibilities

1)
Prediction with hidden layers

2)
Learning neural network

Artificial Neural Networks

Input:

-
From the vehicle changing lanes and the vehicle(s) in the destination
lane

-
Positions (longitudinal and lateral)

-
Velocities

-
Acceleration and / or deceleration


Output:

-
Changes in the positions (longitudinal and lateral) of the vehicle
changing lanes

Artificial Neural Networks

Creating the model


-
Weighted Matrix


-
Populated by the inputs over time


-
Changes constantly

-
Back propagation network


-
Established ANN


-
Used extensively


-
Supervised learning


-
Has error thresholds

Artificial Neural Networks

Ranjeet

Singh
Tomar
* and
Shekhar

Verma

Results:


-
ANNs are practically useless


-
Results are not accurate enough


-
While predictions are not always impossible, more
must rely on non
-
impulsive information instead of
driver behavior

Artificial Neural Networks

Drivers (humans) and vehicles (information)
working collaboratively

-
There are some things that humans/drivers cannot predict or know

-
There are things that vehicles do not know

-
Until full autonomous vehicles, cooperative systems work best

Cooperative
systems

1.
Unseen events ahead

2.
Driving behavior

3.
Possible warnings/alerts and even (some) takeover
from the driver

Cooperative systems


We have the technology and resources to create
these systems



Some infrastructure is already in place



For the most part, savings outweigh costs in the
long run

Intelligent Transportation
Systems

Objects/nodes in system

-
Vehicles

-
Traffic lights

-
Traffic cameras

-
Roadside nodes

-
Roadside servers/computational systems

Intelligent Transportation Systems

I.
Honesty

II.
Security

III.
Accuracy

IV.
Reliability

V.
Over population

VI.
U
nder
population

Major Issues Overall

1.
No system is perfect, yet

2.
Hard to predict humans

3.
Computers aren’t human

4.
Fully autonomous vehicles are
the future

happening
now



CONCLUSION

Neural Network Based Lane Change
Trajectory
Prediction in Autonomous Vehicles

Ranjeet

Singh
Tomar
* and
Shekhar

Verma


Published in Transactions on Computational
Science XIII


2011

Springer Berlin / Heidelberg


http://www.springerlink.com/content/j6p81563n
7302666/fulltext.pdf?MUD=MP

Evaluation of VANET
-
based Advanced
Intelligent Transportation
Systems

Yi Yang,
Rajive

Bagrodia


Published in VANET ’09 Proceedings of
the sixth ACM international workshop on
VehiculAr

InterNETworking



http://dl.acm.org/ft_gateway.cfm?id=16
14273&ftid=685868&dwn=1&CFID=8048
4367&CFTOKEN=49494235

REFERENCES