Robotic Sensor Networks: from theory to practice Sameera Poduri QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. CSSE Annual Research Review 03.17.09 QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. 2/28 oil spill Roomba 3/28 Ecological macroscopes Adaptive sampling QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. Networked Infomechanical systems 4/28 keep warfighters or first responders covered with communications QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. 5/28 Problem Design motion controllers for a robotic sensor network Objectives: 1. communication network is connected 2. sensing coverage is maximized 3. intruder pursuit time is minimized 4. field estimation error is minimized Challenge: global objectives using local sensing and control 6/28 Problem Design motion controllers for a robotic sensor network Objectives: 1. communication network is connected 2. sensing coverage is maximized 3. intruder pursuit time is minimized 4. field estimation error is minimized 7/28 Network Connectivity Given a large network, find local conditions that guarantee global k-connectivity. S. Poduri, S. Pattem, B. Krishnamachari, G. S. Sukhatme. "Using Local Geometry for Tunable Topology Control in Sensor Networks". In 8/28 IEEE Transactions on Mobile Computing, Feb 2009 Neighbor-Every-Theta Condition NET Condition: A neighbor in each θ sector Qu i ck Ti m e ™ an d a TI F F ( LZ W) d ec om pr es so r ar e n ee de d t o s ee t h is pi ct u re . Qu i ck Ti m e ™ an d a TI F F ( LZ W) d ec om pr es so r ar e n ee de d t o s ee t h is pi ct ur e . NET Graph: A graph in which every non-boundary node satisfies NET condition Boundary nodes 9/28 Connectivity of NET graphs Edge connectivity of a NET graph is at least 2 for single parameter, tunable general irregular communication model [Ganesan, et al., UCLA/CSD-TR’02] 10/28 Fi Fi NET obs K cov xi x j 2 (x x ) x x j nbd (i ) i j i j K NET xi x j 2 j NET (i ) (xi x j ) xi x j K obs xi x j 2 (x x ) x x i j obstacle(i) i j j Fi cov distance Virtual force Fi cov Virtual force Potential Fields based Controller Fi NET Fi cov distance Fi cov Fi NET Fi obs x& xi m 11/28 Simulation results 2 / 3 2 / 5 2 / 6 12/28 Robot experiments QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. K. Dantu, P. Goyal, and G. Sukhatme, "Relative Bearing Estimation from Commodity Radios", To appear in IEEE International Conference on Robotics and Automation, Sep 2009 13/28 Minimal sensing ordering information is sufficient to construct a loop [ ] 14/28 Problem Design motion controllers for a robotic sensor network Objectives: 1. communication network is connected 2. sensing coverage is maximized 3. intruder pursuit time is minimized 4. field estimation error is minimized 15/28 Coverage optimization A. Deshpande, S. Poduri, D. Rus and G. S. Sukhatme,”Coverage Control with Location-dependent Sensing Models”, ICRA 2009 16/28 Data-driven approach • pilot deploy at 14 locations • measure sensing coverage • compute optimal locations Uniform deployment 11 cameras Optimized deployment 9 cameras 17/28 Camera Model f p q , p k1 ( p) p q k2 ( p) 2 18/28 Problem Design motion controllers for a robotic sensor network Objectives: 1. communication network is connected 2. sensing coverage is maximized 3. intruder pursuit time is minimized 4. field estimation error is minimized 19/28 Pursuit evasion How should robots move to capture all evaders? M. Vieira, R. Govindan, and G. Sukhatme, "Scalable and Practical Pursuit-Evasion", To appear in International Conference on Robot Communication and Coordination, Mar 2009. 20/28 Setup 32 42 41 40 38 39 6 4 31 37 36 35 34 106 33 5 105 30 3 5 1 29 6 21 20 28 27 56 26 55 25 4 24 23 104 19 18 22 43 3 17 103 101 7 16 15 14 13 12 9 2 11 10 2 1 8 102 48 54 44 8 53 52 7 49 108 47 51 107 45 50 46 #pursuers >> #evaders 1 Stargate 2 tmoteSky 3 MicaZ localization as a service 4th fl. opponent strategy is known same speed 21/28 Results 22/28 Problem Design motion controllers for a robotic sensor network Objectives: 1. communication network is connected 2. sensing coverage is maximized 3. intruder pursuit time is minimized 4. field estimation error is minimized 23/28 Observing marine ecosystems Mapping and sampling of hydrographic features pertinent to aquatic microbial populations 24/28 25/28 Adaptive Sampling Reconstruct a scalar field (temperature, chlorophyll, etc.) Unlike conventional mobile robotics mapping Sensor reading are only valid locally Correlation between sensors decreases rapidly with distance Intuition: the more data near the locations where a field estimate is desired, the less the reconstruction error The spatial distribution of the measurements (the samples) affects the estimation error 26/28 QuickTime™ and a Motion JPEG OpenDML decompressor are needed to see this picture. 27/28 QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. http://robotics.usc.edu/~sameera 28/28 surveillance QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture. 29/28