Intelligent vehicles and road transportation systems (ITS) Week 9 : - - PowerPoint PPT Presentation

1-févr.-15 1

Intelligent vehicles and road transportation systems (ITS)

Week 9 : Multi-vehicle cooperative and collaborative scenarios

ME470

Denis Gingras January 2015

D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 2

Course outline

 Week 1 : Introduction to intelligent vehicles, context, applications and

motivations

 Week 2 : Vehicle dynamics and vehicle modelling  Week 3: Positioning and navigation systems and sensors  Week 4: Vehicular perception and map building  Week 5 : Multi-sensor data fusion techniques  Week 6 : Object detection, recognition and tracking  Week 7: ADAS systems and vehicular control  Week 8 : VANETS and connected vehicles  Week 9 : Multi-vehicular cooperative and collaborative scenarios  Week 10 : The future: toward autonomous vehicles and automated driving

(Final exam)

D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 3

 Brainstorming and introduction  Context and motivation for vehicular cooperation and collaboration  Cooperation for improved positioning and navigation  Cooperation for extended perception  Collaboration for traffic optimization  Collaboration for platooning  Centralized architectures  Decentralized architectures  How to select the proper information to exchange?  The over-convergence problem  The deadlock problem  Some safety examples in highway and urban scenarios  Cooperation and collaboration for driving automation

Week 9 outline

D Gingras – ME470 IV course CalPoly Week 9

4

Why considering multi vehicle architectures?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 4 D Gingras – ME470 IV course CalPoly Week 9

5

What is collaboration?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 5 D Gingras – ME470 IV course CalPoly Week 9

6

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 6 D Gingras – ME470 IV course CalPoly Week 9

What is cooperation?

7

What do we need to achieve cooperation and collaboration?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 7 D Gingras – ME470 IV course CalPoly Week 9

8

For what kind of applications do we need cooperation?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 8 D Gingras – ME470 IV course CalPoly Week 9

9

For what kind of applications do we need collaboration?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 9 D Gingras – ME470 IV course CalPoly Week 9

10

What are the advantages of multi-vehicle approaches

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 10 D Gingras – ME470 IV course CalPoly Week 9

11

What are the drawbacks of multi-vehicle cooperative approaches

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 11 D Gingras – ME470 IV course CalPoly Week 9

12

What would be some possible criteria for priority assignment in vehicular data exchange ?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 12 D Gingras – ME470 IV course CalPoly Week 9

13

What are the parameters that define a group of vehicles?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 13 D Gingras – ME470 IV course CalPoly Week 9

14

What are the differences between a centralized and a decentralized approach in multi-vehicular scenarios?

Brainstorming

Open questions and introductory discussion

Brainstorming 1-févr.-15 14 D Gingras – ME470 IV course CalPoly Week 9

Motivation for cooperation

 Single vehicle solutions do not account for extra positional information

available from surrounding vehicles

 Cooperation among vehicles has the potential to:



Improve positioning/localization accuracy & reliability of each vehicles



Enhance the range of perception (extended environment map )



Mitigate occlusion problems



Allow better prediction of communicating vehicles



Improve quality of navigation information and map matching



Mitigate variability in signal characteristics and environmental conditions

Introduction

Introduction 1-févr.-15 15 D Gingras – ME470 IV course CalPoly Week 9

Motivation

 Cooperative vehicles have the potential to better perform tasks than

isolated single vehicle because they

 Share complementary information (improve accuracy)  Share redundant information (improve robustness/reliability)

 In principle, more information about a phenomenon can be gathered

from multiple measurements

 Exploit multiple sensors within vehicles (GPS, inertial, etc.)  Cluster of vehicles seen as multiple sources of information

 Limited local information gathered by a single vehicle requires

collaboration to resolve inconsistencies between measurements, such as those due to malfunctioning sensors (ex. loss of GPS signal).

 The attractiveness of cooperation in ad-hoc networks lies in its

independence from any major additional infrastructure other than the vehicular communication systems.

1-févr.-15 16 D Gingras – ME470 IV course CalPoly Week 9

Introduction

Introduction

Assumptions

Introduction

Introduction 1-févr.-15 17 D Gingras – ME470 IV course CalPoly Week 9



Vehicles are seldomly alone



Vehicles have sensing, computing and communicating capabilities



Vehicles are “made aware” of surrounding vehicles (relative position) and local environment

 Cooperative approaches combine information from multiple

vehicles to construct a larger, more accurate environmental map

• f their surrounding, including their respective absolute and

relative positions, beyond what is possible from a single vehicle.

1-févr.-15 18 D Gingras – ME470 IV course CalPoly Week 9

Typical collaborative system main components. Note: collaboration

• ften implies

cooperation as well as indicated here.

Source: Baber J et al., Collaborative Autonomous Driving, Intelligent Vehicles Sharing City Roads, IEEE Robotics & Automation Magazine, 2005

Introduction

Introduction

1-févr.-15 19 D Gingras – ME470 IV course CalPoly Week 9

Effects of Vehicle-Infrastructure Cooperative Systems that support driving

Source: Toyota

support driving and aim to prevent traffic accidents by notifying drivers of information, which cannot be detected by a vehicle's own sensors, through communications between vehicles and infrastructure, or among vehicles.

Gain in safety with Cooperative ITS

Introduction

Introduction

1-févr.-15 20 D Gingras – ME470 IV course CalPoly Week 9

Source: Redding J. et al., An Intelligent Cooperative Control Architecture,

An intelligent Cooperative Control Architecture

Gain in safety with Cooperative ITS

Introduction

Introduction

1-févr.-15 21 D Gingras – ME470 IV course CalPoly Week 9

 In CVS systems, vehicles broadcast their physical state information

• ver a shared wireless network to allow their neighbors to track them

and predict possible collisions.  The physical dynamics of vehicle movement and the required accuracy from tracking process dictate certain load on the communication network.  The network performance is directly affected by the amount of offered load, and in turn directly affects the tracking process and its required load.  The tight mutual dependence of physical dynamics of vehicle (physical component), estimation/tracking process and communication process (cyber components) require a new look at how such systems are designed and operated.

Source: Toyota

Tight coupling in cooperative vehicle safety (CVS)

Introduction

Introduction

 A typical cooperative architecture consist of the following tasks:



Detect nearby vehicles in a given range;



Determine cluster topology and vehicle membership;



Estimate and register absolute position (local data fusion) of each vehicle member in the same reference coordinate frame;



Estimate inter-vehicular distances, heading (relative positions) of member vehicles;



Compute confidence interval on all local estimates: measures the accuracy/uncertainty of the local absolute/relative position of estimates;



Select relevant local estimates and uncertainty data for broadcasting to vehicle members (data positioning integrity);



Register remote data with local data in each vehicle (in time and space)



Include remote data to local data in fusion systems in order to perform global fusion and improve local estimates (position or perception information) of individual vehicle members.

Design of cooperative multi-vehicular system

Introduction

Introduction 1-févr.-15 22 D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 23 D Gingras – ME470 IV course CalPoly Week 9

Source: Lin S-p et al., A Multiple Stack Architecture for Intelligent Vehicles, IEEE Intelligent Vehicles Symposium (IV) June 8-11, 2014. Dearborn, Michigan, USA , 2014

Multiple stack layered architecture for collaborative driving system

Introduction

Introduction

1-févr.-15 24 D Gingras – ME470 IV course CalPoly Week 9

Introduction

Introduction

Source: SrwI Texas

Emergency response vehicles with right-of-way, such as ambulances or fire trucks, could alert vehicles nearing an intersection that they were approaching, to more safely and efficiently pass through traffic.

1-févr.-15 25 D Gingras – ME470 IV course CalPoly Week 9

Introduction

Introduction

Source: SrwI Texas

V2V to alert approaching vehicles of traffic situations, such as in an emergency braking scenarios, to help prevent secondary incidents that

• ften occur.

1-févr.-15 26 D Gingras – ME470 IV course CalPoly Week 9

Cooperative perception

Cooperative perception

Source:

Each vehicle share information with nearby vehicles about its own perceived environment, thus allowing to build an extended map.

1-févr.-15 27 D Gingras – ME470 IV course CalPoly Week 9

Cooperative perception

Cooperative perception

The top two figures are the local

• ccupancy grid maps built by

two vehicles. The bottom-left figure shows the merging effect using a low-accuracy GPS based localization results. There is a large alignment inconsistency between the two local maps. The bottom-right figure shows the results using an occupancy grid maps merging method. In this case, the two local maps are properly aligned, leading to consistent results.

Source: Li H. et al, A New Method for Occupancy Grid Maps Merging: Application to Multi-vehicle Cooperative Local Mapping and Moving Object Detection, 12th Int. Conf on Control, Automation, Robotics & Vision, Guangzhou, China, 2012

1-févr.-15 28 D Gingras – ME470 IV course CalPoly Week 9

Model for chain of car collisions. Car is out of the driver’s field of view, but using vehicle-

to-vehicle communication, the driver can be forewarned.

Cooperative perception

Cooperative perception

 Cooperative positioning enable vehicles with the best navigation sensors

to localize vehicle with poorer position estimates.

 From the 2x2 covariance matrices of the position estimates (in a 2D

plane), we obtain the 2 main axis of the confidence ellipses (PCA of the covariance matrix).

 Once the data from the cooperating vehicles are added into the

“extended” state vectors and covariance matrices, the decentralized multi- vehicle data fusion prediction and update recursive/iterative equations take on the same general form as the single vehicle algorithm.

 Collaboration and cooperation implies coherence and synchronization.

Therefore shared information must be registered both in time (time stamped) and in space (same reference frame). Also the shared information must be representing the same quantities and using the same format protocol.

 Standards are therefore imperative for proper data exchanges.

Multi-vehicle collaborative positioning

1-févr.-15 30 D Gingras – ME470 IV course CalPoly Week 9

Source:

 Solution to the cooperative position and pose estimation problem of vehicles in ad-hoc vehicle networks are usually based on decentralized fusion filters.  Specifically, a decentralized filter operating in a cooperative federated structure for enhancing the estimation accuracy of each vehicles states over wireless communication networks subject to uncertain and limited measurements. The filter relies on a variety of position measurements obtained from:  on-board vehicle positioning system,  other cooperating vehicles in the vicinity,  immediate roadside environment via communication.

Cooperative positioning & localization

Cooperative positioning

Motivation for cooperation in positioning

V to V Range

Wireless Access Point Wireless Access Point Single-vehicle based initial position estimate uncertainty (yellow) Cooperative position estimate uncertainty (orange)

Collision risk

Lane ambiguity

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 31 D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 32 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

 White vehicle measure pseudoranges and estimate its GPS position with its covariance matrix (red ellipse).  White vehicle applies its own road map constraints (white lines) to its GPS position (white hatch) in (a).  White vehicle uses the received pseudoranges from the blue vehicle and computes the GPS position and corresponding position covariance of the blue vehicle (tight coupling).  Then the white vehicle applies the road constraints of the blue vehicle (blue lines) to its own position (b). Then white vehicle uses the road map constraints of the black vehicle (c) etc.

Improving ego-GPS position using map constraints from nearby vehicles

Source: Rohani M, Gingras D et al., Vehicular Cooperative Map Matching, IEEE Int. Conf on Connected Vehicles (ICCVE), 2014, Vienna, Austria.

1-févr.-15 33 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

Position estimate improvement using cooperation and map constraints.

Improving ego-GPS position using map constraints from nearby vehicles

Source: Rohani M, Gingras D et al., Vehicular Cooperative Map Matching, IEEE Int. Conf on Connected Vehicles (ICCVE), 2014, Vienna, Austria.

 Time is space in vehicular technology.  Problem of real time sensor fusion with spatially and temporally

misaligned dissimilar sensors.

 Spatial–temporal registration model for intra and inter-vehicular

sensors including radar, lidar, GPS (global positioning system), inertial navigation system (INS), and camera is required for sensor alignment.

 Recursive filters are used to fuse and register simultaneously the

sensors that are locally installed on the vehicles of the cluster.

 When the road geometry information is available from a digital map

database, constrained data fusion schemes are used to improve the fusion accuracy.

 Some sensor self-diagnosis is required: the accuracy of the vehicle-

state estimation is too low otherwise to meet the requirements of collaborative positioning of multiple vehicles.

Simultaneous registration and fusion of multiple dissimilar sensors

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 34 D Gingras – ME470 IV course CalPoly Week 9

 For some safety applications, position estimate resolution and accuracy must

be better than 1 meter. It is dependent of context and traffic density.

 Communication hops and

broadcasting frequency will also depend on traffic density and average speed.

Positioning accuracy requirement in cooperative multi-vehicle schemes

1-févr.-15 35 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

V0

Motion

V0 V4

• bservations

V2 V3 V1

Time stamp=T1 Time stamp=T0

Propagation of uncertainties in cooperative position estimation

The confidence ellipses are an ideal upper bound of the real current error (the difference between the real position and the computed position). In practice, the real error is unknown and the positioning system is only able to calculate the confidence ellipses. Level of confidence is represented by ellipses whose centres are the best estimation of the position.

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 36 D Gingras – ME470 IV course CalPoly Week 9

 Method based on geometric

analysis of sensing uncertainty and used to evaluate the propagation

• f uncertainty in multi-sensor data

fusion techniques .

 Motivated by the idea that the

volume of uncertainty ellipsoids should be minimized by adding information from other sources.

 Example: Say vehicle 2 is the one

with higher position uncertainty (has a large error variance). The

• ther two vehicles have a more

reliable position estimates.

Vehicle 1 GPS Satellite constellation Good LOS Bad LOS Vehicle 2 Vehicle 3

Geometric data fusion

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 37 D Gingras – ME470 IV course CalPoly Week 9

 Exploits useful inter-vehicle data to extract good position measurements

from vehicles with good GPS satellite LOS (line of sight), in order to enhance low positioning accuracy of other vehicles

Uncertainty

ellipse of Vehicle 1 Uncertainty ellipse of Vehicle 2 Uncertainty ellipse

• f vehicle 3

Final uncertainty

• f Vehicle 2

Vehicle 1 GPS Satellite constellation Good LOS Bad LOS Vehicle 2 Vehicle 3

Cooperative positioning & localization

Cooperative positioning

Geometric data fusion

1-févr.-15 38 D Gingras – ME470 IV course CalPoly Week 9

 Select vehicles to collaborate so as to  Maximize information return  While minimizing  Latency  Bandwidth consumption  Computational burden

 Routing protocol to automatically direct a vehicle query into regions of high information content (information utility measure)  Tradeoff between maximum information gain and minimum transmission cost

Posterior from V1 Posterior from V2 “Big” improvement! “Small” improvement! Vehicle V1 Vehicle V2 Current position estimate

Information Driven Vehicle Querying

By identifying vehicles with high position accuracy, we can use their location estimates to help better localize neighbor vehicles with lower location accuracy.

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 39 D Gingras – ME470 IV course CalPoly Week 9

 Putting together all available information from onboard and off-board

systems in a cooperation system is quite a difficult task if the final goal is to maintain constant accuracy and reliability despite the highly dynamic evolution of the vehicle cluster and the environment.

 Intractable performance: The required performance cannot be 100%

guaranteed for all scenarios. Requires statistical approaches.

 Collaborative positioning systems have to be able to autonomously qualify its

• utputs and maintain position integrity.

 Positioning integrity can be interpreted as the aptitude to detect and then

eliminate aberrant measurements in order to estimate a position whose confidence (inaccuracy) are quantified.

 Confidence can be defined as the probability associated with resulting

position estimate being valid, under the assumptions considered.

 Sensor Management is required to ensure sensor data is formatted and

processed in a timely and accurate manner in order to improve estimation and prediction of given vehicles’ position.

Design issues

1-févr.-15 40 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

 In cooperative positioning, each vehicle measures its position and

the relative distance to nearby vehicles.

 In data fusion schemes, positioning data are exchanges between

vehicles and are used to estimate their own position.

 The received measurements data suffers from various time-delays.

The data must be transmitted with time-stamps and the receiver must estimate the time delays due to communication channels.

 Measurements sent from farther cars are received with larger time-

delay depending on the number of hops. It follows that accuracy of the estimates of farther cars become less reliable.

 Hence, only states estimates (attributes) of nearby cars are usually

taken in priority to reduce uncertainty and reduce computing/ communication effort.

Considerations on communication delays

1-févr.-15 41 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

 Similar approach as for single vehicle but with more degrees of freedom as we

include the states of all vehicle in the cluster.

 A state – space description of the global cluster model) is used  A global state vector (multiple vehicles), , is created describing the

dynamics and position of all vehicles in the cluster.

 The global state vector is composed of



The individual self and nearby vehicles state sub-vectors



The environment fixed features sub-vectors (if present)

 All information is time-stamped and registered in the same reference frame  Data fusion filters are used for up-dating position estimates

(prediction/correction) and uncertainty covariance matrix of the estimated states,

 Communication protocols are used to share “good” estimates between local

vehicles.

Centralized cooperative positioning approach ( ) k x

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 42 D Gingras – ME470 IV course CalPoly Week 9

 In the limit (for a given position), as the number of observations

increases, if each vehicle directly observes its collaborators, all

• f the vehicle and feature estimates become completely

correlated with each other.

 Multiple vehicles performing positioning together can attain a

lower absolute error than the single vehicle initial covariance

 The full correlation property of single vehicle positioning scales

to the collaborative positioning case, as the neighbor vehicles are, in essence, extra moving sensors in the state space model.

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 43 D Gingras – ME470 IV course CalPoly Week 9

 Given a vehicle Vi in a cluster of n vehicles .  Having several vehicles moving on a road. The objective is to estimate

the position of each car. At time stamp k, Each vehicle measures not

• nly their absolute position xi , but also the relative (distance)

positions to neighbor vehicles. Each vehicle has his current absolute and relative position estimates given by a state vector.

Vi … Vi+1

…

Vi-1

1 i

D 

1 i

D 

i j

D 

Problem statement (longitudinal motion 1D case)

1-févr.-15 44 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

 Shared data can be fused to compensate for poor satellite

navigation precision in some areas  Triangulation resources are thus made available for position estimation. In a

nutshell, vehicles could behave as navigation satellites for each other

 Provided known positions of different vehicles, establish position of another

using data fusion techniques and geometric fusion.

V2 V3 V4 V1

1

ˆ d

2

ˆ d

3

ˆ d

1 p i i i

x W x



 

   

1 1 1 1

T T

p i i i i i i i i

W J Q J J Q J

   

          



where is the covariance matrix of is a vector of positions and distances W is the optimal weighting matrix minimising uncertainty

T

i i i i

J Q J x x

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 45 D Gingras – ME470 IV course CalPoly Week 9

 In centralized architecture cooperative positioning, in the limit as the

number of observations increases, the lower bound on the covariance matrix of any vehicle becomes equal to the sum of the collaborating vehicles inverses covariance matrices at the time of the initial

• bservation of those vehicles.

 Assuming independance of the 2n vehicles, it can been shown that the

inverse of the covariance matrix of the vehicle i state vector estimate becomes (see Fenwick paper in additionnal readings Week 9),

1 1 1 1 2 2 2 2

lim lim , 1,2,3,... ...

i

V V j k k V V V V V V V V

k k k k j n k k k k

          

                            

i i i i i i i i i

• 1
• 1
• 1
• 1
• 1
• 1

P P P P + P + P

Uncertainty reduction in centralized cooperative positioning

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 46 D Gingras – ME470 IV course CalPoly Week 9

Last equation makes sense in the context of conservation of information. In the general case, P -1 represents the amount of information (in the Fisher sense) present in the system (see Maybeck, 1976). The total amount of information in the system can never decrease, but can stay constant when no noise is added to the system. The sum of information present initially in the system is equal to the inverse of the sum of initial uncorrelated vehicle position errors The amount of information present after infinite observations are made is encapsulated in a single vehicle position covariance.

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 47 D Gingras – ME470 IV course CalPoly Week 9

 The vehicle covariance matrices for both vehicles become fully

correlated and thus identical as time goes to infinity.

 In practice, it rarely happens because the vehicles are moving and a

very limited number of measurements are available for any given position.  For the case of a collaboration between two vehicles. The lower performance bound becomes:

   

lim lim ,

1 2

V V k k V V

k k k k

 

          

1 2

• 1
• 1
• 1
• 1

P P P P

Uncertainty reduction in collaborative positioning

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 48 D Gingras – ME470 IV course CalPoly Week 9

 Decentralised Data Fusion has many advantages

 Removes bottleneck and risk factors associated with centralized

systems

 Distributes processing and communication across several vehicles  Data fusion which occurs at each vehicle is based on its own

information source and from the information generated from surrounding vehicles

 No vehicle forms a global data fusion of the total information at once.  Global solution can be achieved if the decentralized fusion is in a

broadcast mode and all vehicles can communicate with all others

 Scalability of the whole system due to the removal of limitations on

processing power and bandwidth

 Robustness of the system when one node (vehicle) fails.  Modularity, since each vehicle does not require knowledge of the

network topology.

Centralized vs Decentralised Data Fusion

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 49 D Gingras – ME470 IV course CalPoly Week 9

 Fusion based on

• f Bayes rules

 Fusion based on fuzzy

logic rules

Global position correction error (decision fusion) schemes in localization systems

Source: Wann Chin-Der, Lin Ming-Hui, 2004

Cooperative positioning & localization

Cooperative positioning 1-févr.-15 50 D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 51 D Gingras – ME470 IV course CalPoly Week 9

IN each vehicle, the structure of federated distributed data fusion involves several local filters (LFs) and a master fusion filter (MF). There is a LF dedicated to each particular

• sensor. Sensory

data are received fro other vehicles.

Cooperative positioning & localization

Cooperative positioning

Source: Edelmayer A. et al., Cooperative federated filtering approach for enhanced position estimation and sensor fault tolerance in ad-hoc vehicle networks, IET Intelligent Transport Systems, 15th World Congress on ITS.

1-févr.-15 52 D Gingras – ME470 IV course CalPoly Week 9

The delta blocks provide residuals to detect changes in the locally filtered data and trigger actions for filter reconfiguration in the global fusion process.

Cooperative positioning & localization

Cooperative positioning

Cooperative centralized architecture involving N

• vehicles. Data fusion filters

are distributed while sensory and fused data are transmitted between vehicles.

Source: Edelmayer A. et al., Cooperative federated filtering approach for enhanced position estimation and sensor fault tolerance in ad-hoc vehicle networks, IET Intelligent Transport Systems, 15th World Congress on ITS.

1-févr.-15 53 D Gingras – ME470 IV course CalPoly Week 9

Cooperative positioning & localization

Cooperative positioning

Structure and operation

• f a typical centralized

architecture from the point of view of the ego-vehicle (vehicle 1) using distributed filters, that is, the other vehicles share sensory data a as well as fused data (MF estimates).

Source: Edelmayer A. et al., Cooperative federated filtering approach for enhanced position estimation and sensor fault tolerance in ad-hoc vehicle networks, IET Intelligent Transport Systems, 15th World Congress on ITS.

1-févr.-15 54 D Gingras – ME470 IV course CalPoly Week 9

 Traditional centralised filtering solutions cannot cope with the distributed nature of the multi-vehicle dynamic problem.  They are prone to both sensor and communication failures and as such cannot provide satisfactory results either for estimation accuracy or fault tolerance.  Distributed filters among vehicles increase robustness and performance.

Cooperative positioning & localization

Cooperative positioning

Source: Edelmayer A. et al., Cooperative federated filtering approach for enhanced position estimation and sensor fault tolerance in ad-hoc vehicle networks, IET Intelligent Transport Systems, 15th World Congress on ITS.

 Data fusion collaborative architecture

 Sensor data from all slave vehicles are combined  Higher computational burden since higher dimensional data are

jointly processed

 Higher communication burden to transfer sensory data from vehicle

to vehicle.  Decision fusion architecture  Only local estimates and local decisions (hard or soft) based on local computation are transmitted and combined at the master vehicle  Lower computational burden since only low dimensional data (local decisions) are jointly processed at master vehicle  Some communication burden if component decisions/estimation are made at different nodes (vehicles).

Two levels of fusion in centralized collaborative scenarios

Collaborative approaches

Collaboative approaches 1-févr.-15 55 D Gingras – ME470 IV course CalPoly Week 9

 Given some common mission or task (safety or traffic efficiency driven), a multi-vehicle system displays collaborative behavior if, due to some underlying mechanism (i.e., the ’mechanism

• f collaboration’), there is a

concerted and orchestrated behavior of the vehicles to realize this common task.  Still an ongoing research  Most collaborative experiments have been done with low speed vehicles.

1-févr.-15 56 D Gingras – ME470 IV course CalPoly Week 9

Source: Baber J et al., Collaborative Autonomous Driving, Intelligent Vehicles Sharing City Roads, IEEE Robotics & Automation Magazine, 2005

Collaborative approaches

Collaboative approaches

1-févr.-15 57 D Gingras – ME470 IV course CalPoly Week 9

Typical collaborative system architecture

Collaborative approaches

Collaboative approaches

Source: Baber J et al., Collaborative Autonomous Driving, Intelligent Vehicles Sharing City Roads, IEEE Robotics & Automation Magazine, 2005

 Data fusion across multiple sensors within each vehicle is desirable

due to a priori knowledge available on sensor properties, no communication burden and high proprioceptive sensors coherence.

 Attribute/Decision fusion across multiple vehicles is desirable due to

lower communication cost and independence from sensors properties.

 Inconsistencies (conflicts) between different sensors/measurements,

such as due to sensors malfunction, can also be taken into account (ex. Use of belief or fuzzy framework)

 Distributed Attribute/Decision fusion are less sensitive to such

inconsistencies

Data Fusion vs Decision Fusion in collaborative schemes

Collaborative approaches

Collaboative approaches 1-févr.-15 58 D Gingras – ME470 IV course CalPoly Week 9

 Pros

 Maximal exploitation of available complementary information

in multiple sources to improve accuracy/reliability

 Make the localization scheme more robust by exploiting

redondancy of information from similar sources in case of sensor failure (ex. Loss of GPS signal)

 Cons

 High computational burden  High communication burden if data fusion across multiple

vehicles

 Dealing with larger amount of data than decision fusion  Inconsistencies between measurements could cause

performance degradation (malfunctioning sensors, e.g.)

Data fusion vs decision fusion in collaborative scenarios

Collaborative approaches

Collaboative approaches 1-févr.-15 59 D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 60 D Gingras – ME470 IV course CalPoly Week 9

Conclusion

 Cooperative localization systems do not require extensive infrastructure  Collective cluster position accuracy increases with vehicle (node) density  Many architectures (centralized/distributed) are possible;  Decentralized architectures offer more possibilities/robust performance  But decentralized schemes involves higher computational-communication burden  Key is to combine minimum amount of information that yields desired

performance (energy conservation). We have two objective functions:

 Minimize computational burden at each vehicle  Minimize communication burden between vehicles

 Sensors properties and vehicle inter-distances influence level of uncertainty  Some form of cooperative techniques is in general valuable and desirable to yield

reliable performance in sensor networks

 Harsh and variable environmental conditions  Limited capabilities of sensor nodes

 Two main forms of collaborative approaches: decision fusion and data fusion

 Attribute/Decision fusion with data from other vehicles  Data Fusion with sensors/measurements within a single vehicle  Inconsistencies between sensors/vehicles may compromise the performance

gains due to collaboration. Tight registration required.

References

Andrews S., Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communications and Cooperative Driving, Ch 46, Handbook of Int Vehicles, Springer, 2012. Avery P A. et al., Cooperative sensor-sharing vehicle traffic safety system, US Patent 7994902 B2, Southwest Research Institute Texas, 2011. Busson A., Gingras D. et al, Analysis of Inter-vehicle Communication to Reduce Road Crashes, IEEE Transactions On Vehicular Technology, Vol. 60, No. 9, 2011 Dao T S., A Decentralized Approach to Dynamic Collaborative Driving Coordination, PhD Thesis, University of Waterloo, Ont. Canada, 2008. Fenwick J W et al., Cooperative Concurrent Mapping and Localization, Proceedings, IEEE Int. Conf. on Robotics and Automation, ICRA '02, 2002. Halle S. et al., “A Decentralized Approach to Collaborative Driving Coordination,” Proceedings of the 7th IEEE Int. Conf. on ITS, pp. 453 – 458, Washington, D.C, USA, 2004

1-févr.-15 61 D Gingras – ME470 IV course CalPoly Week 9

References

Hulea M et al., A Collaborative Approach to Autonomous Single Intersection Control, 19th Mediterranean Conference on Control and Automation, Corfu, Greece, 2011 Kesting, A., Treiber, M., Schonhof, M., Helbing, D., 2008. Adaptive cruise control design for active congestion avoidance. Transportation Research Part C, (Emerging Technologies), 2007 Li H. et al., “Cooperative multi-vehicle localization using split covariance intersection filter,” in Intelligent Vehicles Symposium (IV), 2012 IEEE, june 2012. Ren W. et al., Distributed Consensus in Multi-vehicle Cooperative Control Theory and Applications, Springer, 2008. SwRI, http://www.swri.org/4org/d10/isd/ivs/cv-lane-level.htm Varaiya, P., Smart cars on smart roads: problems of control. IEEE Transactions on Automatic Control 38 (2), 195–207, 1993. Yang X. et al.,“A Vehicle-to-Vehicle Communication Protocol for Cooperative Collision Warning,” The 1st Annual Int. Conf. on Mobile and Ubiquitous Systems: Networking and Services, MobiQuitous 2004.

1-févr.-15 62 D Gingras – ME470 IV course CalPoly Week 9

1-févr.-15 63

QUESTIONS?

D Gingras - UdeS – IV course CalPoly Week 7