Cooperative Air and Ground
Surveillance
Wenzhe Li
Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

Introduction
The use of robots in surveillance and
exploration is gaining prominence.
Surveillance
 Target detection
 Tracking
 Search and rescue operations

UAV and UGV

UAV(Unmanned aerial vehicle)
Advantage: Move rapidly, Cover large area
Disadvantage: Low accuracy for localization

UGV(Unmanned ground vehicle)
Advantage: High accuracy for localization
Disadvantage: Not move rapidly, can not see through
obstacles.
Main idea arise from answering question :
How to make it both Move rapidly and Accurately
locate target ?
Major topics covered in this paper
In this paper, authors present the approach to
cooperative search, identification, and localization of
targets using a heterogeneous team of fixed-wing UAV
and UGVs.
Three major topics.
Synergy of UAVs and UGVs
 Framework
 Algorithms to search and localization

Contribution of paper
Framework is scalable to multiple
vehicles.
 Decentralized Algorithms for control of
each vehicle
 Easy Implemented, independent of
number of vehicles, offer guarantee for
search and localization

Before moving to next section…
How to integrate UAVs and UGVs ?
 What UAVs and UGVs be responsible
for? (to exhibit complementary capability)
 Why such framework is scalable to large
system?
 What techniques to use to solve
problem?
 ……….

Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

UAV Airframe and Payload
◆ onboard embedded PC
◆ IMU 3DM-G from MicroStrain
◆ external global positioning system (GPS): Superstar GPS receiver from CMC electronics,
10 Hz data
◆ camera DragonFly IEEE-1394 1024 × 768 at 15
frames/s from Point Grey Research
◆ custom-designed camera-IMU Pod includes the
IMU and the camera mounted on the same plate.
The plate is soft mounted on four points inside the
pod. Furthermore, the pan motion of the pod can be
controlled through an external-user PWM port on
the avionics.
Ground Station





Each UAV continuously communicate with
Ground Station Communication : 1hz, up to 6mi
Performs GPS corrections and Flight Update
Concurrently monitor up to ten UAVs
Direct communication between UAVs via
Ground Station and 802.11b
Ground station has an operator interface
program
The UGV Platform
Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

Framework
Information-driven framework
 ASN(Active sensor network) architecture
 Key idea: sensing action -> reduction in
uncertainty
 Utility on robot and sensor state and
actions
 Target Detection
 Target Localization

Screen clipping taken: 2010/3/29, 11:11
Target Detection

Certainty Grid : our representation
certainty grid is a discretestate binary random field in which each element
encodes the probability of the corresponding grid cell being in a particular
state
1. Yd,i(k|k) = logP(x) = logP(s(Ci) = target).
where subscript d denotes detection, stores the accumulated target detection
certainty for cell i at time k
2. id,s(k) = logP(z(k)|x)
Information associated with the likelihood of sensor measurements z
3. Updated by the log-likelihood form of Bayes rule:
Identify cells that have an acceptably high probability of containing features
or targets of interest.
Target Localization
Target Localization : Second part of task
 Problem posed as a linearized Gaussian
estimation problem
 Kalman filter is used

Target Localization






Vector Yf : Coordinates of all the features detected by
the target detection algorithm
Yf,i : denoting the (x, y) coordinates of the feature in a g
lobal coordinate system
Information filter maintains Yf,i(k | k) and matrix
Yf,i(k | k)
Estimation mean and covariance by
Fusion of Ns sensor measurements
Uncertainty Reducing Control





Entropy-based measure
Mutual information measures
Control objective is to reduce estimate uncertainty
Uncertainty directly depends on the system state
and action
Vehicle chooses an action that results in a maximum
increase in utility or the best reduction in the
uncertainty
Scalable Proactive Sensing Network
Can be deployed for searching for targets and for
localization
 Search and localization algorithms are driven by
Information-based utility measures
 Independent of the source of the information
 Nodes automatically reconfigure themselves in this task
 Scales to indefinitely large sensor platform teams

Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

Air-Ground Coordination

The search and localization task consists of two
components:
1. First, detection of an unknown number of
ground features in a specified search area ˆyd
(k|k).
2. The refinement of the location estimates for
each detected feature Yf,i(k|k).
Feature Observation Uncertainty
Optimal Reactive Controller for Localization

Controller is a gradient control law, which automatically
generates sensing trajectories that actively reduce the
uncertainty in feature estimates by solving:
where U is the set of available actions, and If,i(ui(k)) is
the mutual information gain for the feature location
estimates given action ui(k).

For Gaussian error modeling of Nf features
Optimal Reactive Controller for Localization
Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

Aerial images of the test site captured during a typical UAV flyover at 65 m altitude.
Three orange ground features highlighted by white boxes are visible during the pass.
1. When only use UAVs : In excess of 50 passes
(about 80 min of flight time)
2. When only use UGVs : In excess of half an hour
for the ground vehicle
3. When they are collaborative:
completes this task in under 10 min
Outline
Introduction
 Experimental Testbed
 Framework
 Air-Ground coordination
 Experiment Results
 Conclusion

Conclusion
Unique Features:
1. Methodology is transparent to the pecificity
and the identity of the cooperating vehicles.
2. Computations for estimation and
control are decentralized
3. Methodology presented here is scalable
to large numbers of vehicles.