Optimal Policies in Complex Largescale UAS Traffic Management
![Convergence for Utilities Grid cell [3, 4, 4] [3, 2, 4] [2, 1, 1]](https://slidetodoc.com/presentation_image_h/cbbaf2cae8bfde9911175b7feb1b9b94/image-26.jpg "Convergence for Utilities Grid cell [3, 4, 4] [3, 2, 4] [2, 1, 1]")
- Slides: 38
Optimal Policies in Complex Large-scale UAS Traffic Management David Sacharny & Tom Henderson ICPS Taipei, Taiwan 7 May 2019 1 ICPS 2019
Futuristic Vision (Slide from Jared Esselman; UDOT) 2 ICPS 2019
Commercial Use Cases • • • 3 D Mapping, Video Collection Delivery (Amazon, etc. ) Inspections Data (Re)Transmission Air Taxis èInvestment 2017: $506 M è 1000’s of flights per day 3 ICPS 2019
Drone HW Investment ($B) 4 ICPS 2019
Utah Urban Air Mobility Idea (Slide from Jared Esselman; UDOT) 5 ICPS 2019
UDOT UAM (cont’d) (Slide from Jared Esselman; UDOT) 6 ICPS 2019
UDOT UAM (cont’d) Proposal: Airways above roadways. 7 ICPS 2019
UAM: Need to Plan Flights 8 ICPS 2019
Dynamic UAV Flight Path Planning in Urban Environments
Geospatial Intelligence: BRECCIA 10 ICPS 2019
BRECCIA 11 ICPS 2019
URBAN Implementation • • The BRECCIA Agent represents the core abstraction for all agents in the system. Agents are distributed across specialized machines such as UAVs, mobile laptops, or high performance computers. BRECCIA Agent BDI Engine Uncertainty Reduct. Goal P. L. Logic Module Geo. Wave Connector Specialized Functions Hadoop DB Accumulo Geo. Server : Mission Planner RRT* Planner • The inherited components of each BRECCIA agent enable an overall system that is dynamic and datadriven. : Weather Monitor : UAV Manager : User Example Instantiations of the BRECCIA Agent 12 ICPS 2019
Which Path To Take? What about Wind? What about Rain? 13 ICPS 2019
Airway Corridors E. g. , over Salt Lake City
Airspace Volumes Z Y -X X -Y -Z (a) Airspace Volumes (b) Action Directions
Learning Optimal Action Policy 4 x 4 Grid 16 ICPS 2019
Bellman Equations • 17 ICPS 2019
State Representation • 18 ICPS 2019
State Representation: Reduced • 19 ICPS 2019
Actions A = {X, -X, Y, -Y, Z, -Z} * move in one of the coordinate directions 20 ICPS 2019
Probabilistic State Transition 21 ICPS 2019
Reward Function 22 ICPS 2019
Value Iteration Algorithm From: Russell & Norvig 23 ICPS 2019
Experiments • Start Location: 1, 1, 1 (index 1) • Goal Location: 4, 4, 4 (index 64) • Blocked Cell: 4, 4, 3 (index 60) à Can’t exit 4 x 4 x 4 à Preference for horizontal motion 24 ICPS 2019
State Utilities and Path 25 ICPS 2019
Convergence for Utilities Grid cell [3, 4, 4] [3, 2, 4] [2, 1, 1] 26 ICPS 2019
Optimal Policies X: RIGHT Z: UP Y: BACK 27 ICPS 2019
Optimal Policies 28 ICPS 2019
Cell Travel Density 29 ICPS 2019
Policies with Wind in Y No action in Y axis! No Wind 30 Strong Wind in Y Direction ICPS 2019
Current Work: Get Data! 31 ICPS 2019
Current Work: Testing! Deseret UAS 32 ICPS 2019
Conclusions • Developed effective and efficient optimal policy method • Converted core BRECCIA system to work for UAS Traffic Management • allows communicating, autonomous agents • Cloud computing 33 ICPS 2019
Current Development 34 ICPS 2019
Large-scale Simulation • http: //www. cs. utah. edu/~cem/uav/
Questions? 36 ICPS 2019
UTAH UAV Fleet 37 ICPS 2019
Acknowledgment This material is based upon work supported by the Air Force Office of Scientific Research under award number FA 9550 -17 -1 -0077 (DDDAS-based Geospatial Intelligence) 38 ICPS 2019