Decentralized Data Fusion and Control in Active Sensor Networks

15
Decentralized Data Fusion and Control in Active Sensor Networks Alexei Makarenko, Hugh Durrant-Whyte Christian Potthast

description

Decentralized Data Fusion and Control in Active Sensor Networks. Alexei Makarenko , Hugh Durrant -Whyte. Christian Potthast. Motivation. Example I. Example II. Decentralization. Scalable Computational and communication load at each node is independent of the size of the network - PowerPoint PPT Presentation

Transcript of Decentralized Data Fusion and Control in Active Sensor Networks

Page 1: Decentralized Data Fusion and Control in Active Sensor Networks

Decentralized Data Fusion and Control in Active Sensor Networks

Alexei Makarenko, Hugh Durrant-Whyte

Christian Potthast

Page 2: Decentralized Data Fusion and Control in Active Sensor Networks

Motivation

Page 3: Decentralized Data Fusion and Control in Active Sensor Networks

Example I

Page 4: Decentralized Data Fusion and Control in Active Sensor Networks

Example II

Page 5: Decentralized Data Fusion and Control in Active Sensor Networks

Decentralization

• Scalable– Computational and communication load at each node is independent

of the size of the network• Robustness

– No element of the system is mission critical, system is survivable in the event of run-time loss of components

• Modularity– Components can be implemented and deployed independently from

each other

Characterized by:

• No component is central to the successful operation of the network• No central service or facilities

Page 6: Decentralized Data Fusion and Control in Active Sensor Networks

Node structure

Page 7: Decentralized Data Fusion and Control in Active Sensor Networks

Local filter

Page 8: Decentralized Data Fusion and Control in Active Sensor Networks

Local Filter IIEnvironment feature: xk = x(tk)

Observation of feature: zk = z(tk)

Observation likelihood: L(zk | xk)

Find the posterior probability of: P (xk|Zk , x0 )

Prediction of the motion

Fuse the information

Page 9: Decentralized Data Fusion and Control in Active Sensor Networks

Local Filter III

Local belief and the new belief in an external node

Information can be computed as:

Fusing of information held by two different nodes:

Page 10: Decentralized Data Fusion and Control in Active Sensor Networks

IF vs. KF

Page 11: Decentralized Data Fusion and Control in Active Sensor Networks

IF vs. KF

• IF and KF update both in two steps– Prediction and measurement step

• Update steps can vastly differ in complexity– KF prediction step: – IF prediction step:– KF measurement update: – IF measurement update:

O(n2.4 )

O(n2 )

O(n2.4 )

O(n2 )

Page 12: Decentralized Data Fusion and Control in Active Sensor Networks

Control

• Coordinated Control– Chose action purely on local observations– Propagate observed information to sensing platform

• Cooperative Control through Negotiation– Propagate expected information through negotiation

channels.

Page 13: Decentralized Data Fusion and Control in Active Sensor Networks

ExperimentsTracking a target:

Page 14: Decentralized Data Fusion and Control in Active Sensor Networks
Page 15: Decentralized Data Fusion and Control in Active Sensor Networks

Experiments