150 likes | 302 Views
Pursuit Evasion Games (PEGs) Using a Sensor Network. Luca Schenato, Bruno Sinopoli Robotics and Intelligent Machines Laboratory UC Berkeley {sinopoli,lusche}@eecs.berkeley.edu. Outline. Description of the application The role of a sensor network Implementation issues Open problems
E N D
Pursuit Evasion Games (PEGs) Using a Sensor Network Luca Schenato, Bruno Sinopoli Robotics and Intelligent Machines Laboratory UC Berkeley {sinopoli,lusche}@eecs.berkeley.edu
Outline • Description of the application • The role of a sensor network • Implementation issues • Open problems • Tentative roadmap
Current Experimental Setup for PEG • Experiment Setup • -Cooperation of • -One Aerial Pursuer (Ursa Magna 2) • Three Ground Pursuer (Pioneer UGV) • Against One Ground Evader (Pioneer UGV) • (Random or Counter-intelligent Motion) • -Wireless Peer-to-Peer Network Arena: Cell: 1m x 1m Detection: Vision-based or simulated Aerial Pursuer Vehicle Position Vision Sensor Waypt Request Ground Pursuer 3x3m Camera View GroundEvader Vehicle Position Vision Sensor Centralized Ground Station Courtesy of Jin Kim
Current PEG Implementation UAVs Lucent Orinoco (WaveLAN) (Ad Hoc Mode) Ground Monitoring System Ground Mobile Robots Courtesy of Jin Kim
Where does the Sensor Network fit in? UAVs Lucent Orinoco (WaveLAN) (Ad Hoc Mode) Ground Monitoring System Ground Mobile Robots Gateways Sensor Webs Courtesy of Jin Kim
Distributed Pursuit Evasion Games (DPEG) * Robot pictures from ActivMedia website
The role of a sensor network • Provide additional information about evaders’ motion • Relay such information to the pursuers to design and implement an optimal pursue strategy • Possibly provide guidance to pursuers
The general picture • Sensors: • randomly distributed • partial location information • limited communication range and bandwidth, which depends heavily on the topology of the environment • limited computation power • Network: • Ad hoc • Dynamic network topology • Multi hop communication
Implementation Issues • The complexity of the problem suggests an incremental approach to implementation: • Debugging is problematic and costly • Too many things can go wrong at the same time • Extremely difficult to analyze algorithms for the general framework. Divide & Conquer
Implementation Strategy • Implement & test algorithms within TOSSIM • Interface between TOSSIM and Matlab for visualization purposes • Evaluate performances with respect to key objectives: • Accuracy • Power usage • Security • Robustness • Bandwidth efficiency
Approach to experiment • Start with simplified version of the full scale application, i.e. : • Assume motes know their position • Assume robots know their position and move on straight lines at a constant velocity • Debug algorithms on a subnet (<100 nodes) • Add new algorithms as they become available • Develop a monitoring system to track the state of the network • Routing tables , connectivity, data passing etc.
What we need to do in the short term • Let’s do the real thing!!! • Select a big enough space (RFS) • Deploy, test and debug a network of sensors (>=400) • Start with centralized algorithm • Use the test-bed to evaluate algorithms and bootstrap any interesting research projects • Continue with decentralized algorithms
Why starting with centralized approach ? • Algorithms are ready • It will show if, when and why centralized algorithms fail • It will inspire decentralized algorithms • Feasible by January
In the long term: • Interface TOSSIM with a visualization tool and test decentralized algorithms • Implement the most promising on the test-bed (ideally by January)
Would be nice if: • Motes were self programming • There was monitoring system • There was a GUI • There was a smart powering system • There was a “loose” synchronization scheme to avoid clock drift