APPLICATION DOMAIN: SENSOR NETWORKS

Transcription

1 APPLICATION DOMAIN: SENSOR NETWORKS

2 SENSOR NETWORK AS A CONTROL SYSTEM Know nothng - mut deploy reource how many where - Cooperate but operate autonomouly - Manage Communcaton, Energy Data fuon, buld prob. map of target locaton tatc or trajectore dynamc Know everythng - mut deploy reource to mamze beneft from nteractng wth data ource target: trac, get data - Manage Communcaton, Energy COVERAGE Poton/Move to optmze detecton prob. Update event denty nformaton TARGET DETECTION Localze target Poton/Move to optmze data collecton qualty DATA COLLECTION COVERAGE: pertently loo for new target pread node out and earch TRADEOFF: Model for optmzaton Control node locaton to optmze COVERAGE + DATA COLLECTION DATA COLLECTION: optmze data qualty congregate node around nown target Chrto G. Caandra CODES Lab. - Boton Unverty

3 COVERAGE CONTROL Deploy enor to mamze event detecton probablty unnown event locaton event ource may be moble enor may be moble R Hz/m Ω Meguerdchan et al, IEEE INFOCOM, 200 Corte et al, IEEE Tran. on Robotc and Automaton, 2004 Caandra and L, Eur. J. of Control, 2005 Gangul et al, Amercan Control Conf., 2006 Huen and Stpanovc, Amercan Control Conf., 2007 Hoayem et al, Amercan Control Conf., 2007 Perceved event denty data ource over gven regon mon pace Chrto G. Caandra CODES Lab. - Boton Unverty

4 COVERAGE: PROBLEM FORMULATION N moble enor, each located at R 2 Data ource at emt gnal wth energy E Sgnal oberved by enor node at SENSING MODEL: p, P[Detected by A, ] A data ource emt at Senng attenuaton: p, monotoncally decreang n d - R Hz/ 50 m Ω Chrto G. Caandra CODES Lab. - Boton Unverty

5 COVERAGE: PROBLEM FORMULATION Jont detecton prob. aumng enor ndependence [,, N ] : node locaton P, N [ p ], Event enng probablty OBJECTIVE: Determne locaton [,, N ] to mamze total Detecton Probablty: ma R P, d Ω Perceved event denty Chrto G. Caandra CODES Lab. - Boton Unverty

6 CONTINUED DISTRIBUTED COOPERATIVE SCHEME Set [ ] d p R H N N Ω,, Chrto G. Caandra CODES Lab. - Boton Unverty Mamze H,, N by forcng node to move ung gradent nformaton: [ ] d d d p p R H N Ω, H + + β Dered dplacement V Δt

7 CONTINUED DISTRIBUTED COOPERATIVE SCHEME ha to be autonomouly evaluated by each node o a to determne how to move to net poton: CONTINUED Ø Ue truncated p Ω replaced by node neghborhood Ø Dcretze p ung a local grd Chrto G. Caandra CODES Lab. - Boton Unverty Caandra and L, EJC, 2005 [ ] d d d p p R H N Ω, H + + β

8 EXTENSION: POLYGONAL OBSTACLES CONTINUED Contran the navgaton of moble node Interfere wth enng: pˆ, p, f vble from 0 otherwe Vblty Regon Chrto G. Caandra CODES Lab. - Boton Unverty

9 GRADIENT CALCULATION WITH OBSTACLES CONTINUED N pˆ, H R [ pˆ, ] d + V d d, p pˆ 0 vble nvble Q A j j Q: # of occludng corner pont New term capture change n vblty regon of Mathematcally: ue etenon of Lebntz rule for dfferentatng ntegral where both ntegrand and ntegraton doman are functon of the control varable Chrto G. Caandra CODES Lab. - Boton Unverty

10 OPTIMAL COVERAGE WITH OBSTACLES Zhong and Caandra, 2008 Chrto G. Caandra CODES Lab. - Boton Unverty

11 DEMO: OPTIMAL DISTRIBUTED DEPLOYMENT WITH OBSTACLES SIMULATED AND REAL Chrto G. Caandra CODES Lab. - Boton Unverty

12 COVERAGE + DATA COLLECTION Recall tradeoff: COVERAGE: pertently loo for new target pread node out TRADEOFF: Control node locaton to optmze COVERAGE + DATA COLLECTION DATA COLLECTION: optmze data qualty congregate node around nown target MODIFIED DISTRIBUTED OPTIMIZATION OBJECTIVE: collect nfo from detected data ource target whle mantanng a good coverage to detect future event H, t R P, d u D t S u F u, Su : data ource value D t : et of data ource, etmated baed on enor obervaton Fu, : jont data collecton qualty at u e.g., covarance Chrto G. Caandra CODES Lab. - Boton Unverty

13 DEMO: REACTING TO EVENT DETECTION Important to note: There no eternal control caung th behavor. Algorthm nclude tracng functonalty automatcally Chrto G. Caandra CODES Lab. - Boton Unverty