Towards Autoomous ISR by a Team of Coopera;ng Gliders Isaac Kaminer Department of Mechanical and Aerospace Engineering Naval Postgraduate School e-mail: kaminer@nps.edu Joint work with: N. Camacho, B. Carlson, P. Frontera, C. Dillard, S. Tejrson, J. Juriga, V. Dobrokhodov, K. Jones
Quest for Eternal Flight AV Helios: 247 span Breaks up in turbulence off Hawaii, June 26, 2003 AV Global Observer: $120mil JCTD 175 span 55k to 65k, 5-7 days Crashed April 1, 2011
Quest for Eternal Flight Google s Solar-Fueled Cyber Drone Crashes in New Mexico Test Titan Solara 50: 50m wingspan 55k to 65k, months Crashed May 1, 2015 Facebook to begin test flights for Internet-beaming drone Aquila Wingspan of a 737 55k to 65k, months Crashed?
Quest for Eternal Flight QinetiQ: Wingspan of 18 meters but weighing just 30 kg Flew above the clouds (70000 ) for 14 days straight Propeller driven by lithium-sulphur batteries powered by nothing but sunlight (amorphous silicon arrays) Unmanned but RPV, manned launched (5 people) Record flight in YPG, AZ from July 9 to July 23, 2010
Alterna;ve Approaches Two directions in the academic community Solar (ETH Zurich) Thermal Soaring (NASA, NRL, NPS, PennState.) Combine solar and wind energy harvesting enable extended endurance flight of a single platform Replace one by many no catastrophic failures improve robustness of soaring flight Primary focus on extended endurance and range Challenge posed to NPS students: use a team of gliders for an ISR type mission
Towards ISR missions at NPS Two courses in MAE Autonomous Systems: 4811 and 4823 ME4811 Aircraft dynamics System ID Control and Estimation: Autopilot design Thermal detection and tracking Condor: High Fidelity Glider Simulator ME4823 Optimal Control Trajectory generation Search Theory Bayesian Updates Nonlinear Control Path Following State Machines Mission Supervisor Final Project Autonomous Road Patrol by a Team of Three Soaring Gliders MS Thesis
Goal Maintain continuous visual surveillance of a roadway utilizing multiple gliders Problem Description Design Objectives Minimize operator interaction Employ three autonomous gliders to conduct the surveillance One glider patrolling Two gliders searching for energy sources (thermals) 7
Persistent, Autonomous Surveillance General requirements for persistent surveillance Coordination: Gliders sharing information for efficient thermal search Command: Decentralized, flexible assignment to tasks Control: Real-time trajectory planning and following 8
Thermal Search: Encapsula;ng Prior Knowledge IR Image => 3D elevation map PDF from slope slope Bayesian inference is the mechanism of integrating prior knowledge with current measurements
Thermal Search: Coopera;on Each glider is an imperfect sensor False alarm α establishes that the thermal is present when it is not Missed detection β given when the sensor recognizes that the thermal is absent when it is present p(present detected) = (1 β) p 0 α(1 p 0 ) + p 0 (1 β) Bayesian inference updates Defines posterior probability of thermal presence, given an observation Information is shared between all gliders 10
Command One glider must be patrolling the roadway Remaining gliders: conduct search operations Help state gain altitude from thermals discovered If patrolling glider s altitude drops below a defined threshold, it must be relieved on station Decisions made by mission supervisor 11
Control Trajectory Generation Patrol glider s camera must follow the road within space and time requirements Waypoint type trajectory is too coarse Need to generate a suboptimal trajectoryin realtime Sub-Optimal Trajectory minimizes a cost function consisting of: Total distance traveled by the camera footprint along road Surveillance camera rotations in pan and tilt Penalties for exceeding camera limits 12
Control Trajectory Following Once the flight trajectory has been defined to follow it Must minimize error in: Distance from the road Desired heading on the road Serret-Frenet frame utilized Takes tangent to a rabbit point on the road and defines a normal unit vector orthogonal to it 13
Simula;on Environment Simulation conducted using Condor Flight Simulation software Professional glider pilot training tool Provides state information at 100 Hz Accepts commands to the glider via an API
Simulation Big Picture Initialize simulation with three gliders (1, 2, 3) Glider 1 assigned to patrol the road Glider 2 & 3 assigned to search for thermals Glider 1 begins patrol By first generating an online trajectory to patrol the road Glider s 2 and 3 fly to the first waypoint of their search pattern They begin their search for thermals Glider 1 s altitude falls below a threshold A help signal is sent to Glider s 2 and 3 Glider 2 or 3 is selected as the best choice. That glider generates online trajectory to patrol the road Glider 2 or 3 is on station patrolling the road Glider 1 flies to the most cost efficient thermal in the group history 15
Two Approaches Group A Prior probability map determined by accumulation of previous runs Searching gliders conduct lawnmower patterns in pursuit of complete coverage of the area 16
Two Approaches Group B Prior probability map determined by analysis of topography. Slope < 3 (valleys) Slope > 30 (hills) Searching gliders conduct traveling salesman path, optimized to visit potential thermal locations 17
Coordinated Road Search :: Results (Group 1) SIM PARAMETERS Grid Size: 20km x 20km Speed: 25m/s Initial Elevation: 2000m Sink Rate: 0.7m/s (avg) Help Signal: 1300m (16mins) 18
Coordinated Road Search :: Results (Group 2) SIM PARAMETERS Grid Size: 20km x 20km Speed: 25m/s Initial Elevation: 2000m Sink Rate: 0.7m/s (avg) Help Signal: 1300m (16mins)
Results Group A Lawnmower Group 2hrs 24mins of continuous road coverage 3 handoffs Group B Topography Group 1hr 29mins of continuous road coverage 2 handoffs 20
Does it Pay to Have More Gliders? Geographic Average Mission Time Historical Average Mission Time 5 UAVs 5 UAVs 4 UAVs 4 UAVs 3 UAVs AVG Patrol Time AVG Total Time 3 UAVs AVG Patrol Time AVG Total Time 2 UAVs 2 UAVs - 50.0000 100.0000 150.0000 Time - 50.0000 100.0000 150.0000 Time Patrolling Time 160 140 120 100 80 60 40 20 0 Average Patrolling Time 0 50 100 150 200 Total Time 2 UAVs 3 UAVs 4 UAVs 5 UAVs 21
Conclusions Successful integra;on of new key technologies in the AS curriculum Students provided a solu;on to a problem not yet addressed by the research community ISR mission: Coopera;ve road search Coopera;ve thermaling prior informa;on on thermal ac;vity Bayesian inference Decentralized control algorithms that consider low-level centering guidance and high-level mission planning and execu;on strategies; A conference paper submited to ACC 2016 TPS is interested in a course based on this work