Autonomous Flight Systems Laboratory Aeronautics Astronautics All slides
Autonomous Flight Systems Laboratory Aeronautics & Astronautics All slides and material copyright of University of Washington Autonomous Flight Systems Laboratory
Autonomous Flight Systems Laboratory Aeronautics & Astronautics Research and Development at the Autonomous Flight Systems Laboratory University of Washington Seattle, WA Guggenheim 109, AERB 214 (206) 543 -7748 http: //www. aa. washington. edu/research/afsl
Real Time Strategic Mission Planning Autonomous Flight Systems Laboratory Aeronautics & Astronautics Pattern hold/Team assembly Transition Obstacle/Threat Avoidance Base Coordination w/ surface vehicles Searching/Target ID University of Washington 3
System Overview Autonomous Flight Systems Laboratory Aeronautics & Astronautics Previously funded by DARPA & AFOSR University of Washington 4
System Block Diagram Autonomous Flight Systems Laboratory Aeronautics & Astronautics Solving optimal control problems in real-time University of Washington 5
Stochastic Problem Formulation Autonomous Flight Systems Laboratory Aeronautics & Astronautics n Predicted probability of survival of each vehicle n Predicted probability that a task is not completed n Team utility function University of Washington at time tq+1 6
Distributed Architecture for Coordination of Autonomous Vehicles Autonomous Flight Systems Laboratory n n Aeronautics & Astronautics Each vehicle plans its own path and makes task trading decisions to maximize the team utility function There is one active coordinator agent at a time efficiency n failure detection n local/global information exchanges n n n Computational requirement for running coordinator agent is small compared to planning Coordinator role can be transferred to another vehicle via a voting procedure University of Washington 7
Evolution-based Cooperative Planning System (ECo. PS) Autonomous Flight Systems Laboratory n n Aeronautics & Astronautics Uses Evolutionary Computationbased techniques in the optimization of trading decision making and path planning Task planner uses price and shared information in addition to predicted states of the world for making trading decisions Task planner interacts with path planner and state predictor to simultaneously search feasible near-optimal task and path plans. We call this system the “Evolution. Based Collaborative Planning System” – ECo. PS, combining market based techniques with evolutionary computation (EC). University of Washington 8
Evolutionary Computation (EC) Autonomous Flight Systems Laboratory n n Aeronautics & Astronautics Motivated by evolution process found in nature Population-based stochastic optimization technique University of Washington Metaphor Mapping 9
Features of Evolution-Based Computation Autonomous Flight Systems Laboratory n Provides a feasible solution at any time n Optimality is a bonus n Dynamic replanning n Non-linear performance function n Collision avoidance n Constraints on vehicle capabilities n Handling loss of vehicles n Operating in uncertain dynamic environments n Timing constraints University of Washington Aeronautics & Astronautics 10
Market-based Planning for Coordinating Team Tasks Autonomous Flight Systems Laboratory Distributed Task Planning Algorithm Aeronautics & Astronautics Task allocation problem: At trading round n Each vehicle proposes which are approved by the auctioneer based on bid price. At the end of the trading round: The goal of task trading: University of Washington 11
Dynamic Path Planning Autonomous Flight Systems Laboratory Aeronautics & Astronautics n Generate feasible paths and planned actions within a specified time limit (ΔTs ) while the vehicles are in motion. n Highly dynamic environment requires a high bandwidth planning system (i. e. small ΔTs). Formulate the problem as a Model-based Predictive Control (MPC) problem n University of Washington 12
EC-Based Path Planning Autonomous Flight Systems Laboratory Aeronautics & Astronautics Path Encoding Dynamic Planning Mutation University of Washington 13
Collision Avoidance Autonomous Flight Systems Laboratory n n Aeronautics & Astronautics Model each site in the environment as a uncertainty circular area with radius Probability of intersection: use numerical approximation n computationally easier than true solution n : possible intersection region : probability density field function : position on the path Ci : expected site location v : velocity of the vehicle University of Washington 14
Collision Avoidance Example Autonomous Flight Systems Laboratory Aeronautics & Astronautics University of Washington 15
Simulation Results Autonomous Flight Systems Laboratory Aeronautics & Astronautics Simulation on the Boeing Open Experimental Platform University of Washington 16
Some Aspects of ECo. PS Autonomous Flight Systems Laboratory n n n Aeronautics & Astronautics Each vehicle computes its own trajectory and makes decision to trade its tasks with other vehicles. Vehicles may sacrifice themselves if that benefits the team. Each vehicle needs to have periodically updated locations of nearby vehicles only for collision avoidance. Each vehicle needs to know the information about the environment. The accuracy of the information affects the quality of its decision making. The rate of environment information updates should be selected based on how fast objects move in the environment. Assuming vehicles are equipped with on-board sensors, sharing sensed data improves the performance of the team. University of Washington 17
Contact Us Autonomous Flight Systems Laboratory Aeronautics & Astronautics Investigators Dr. Rolf Rysdyk Dr. Uy-Loi Ly Dr. Juris Vagners Dr. Kristi Morgansen Dr. Anawat Pongpunwattana rysdyk@aa. washington. edu ly@aa. washington. edu vagners@aa. washington. edu morgansen@aa. washington. edu anawatp@u. washington. edu Autonomous Flight Systems Laboratory Guggenheim 109 (206) 543 -7748 http: //www. aa. washington. edu/research/afsl Nonlinear Dynamics and Control Laboratory AERB 120 (206) 685 -1530 http: //vger. aa. washington. edu University of Washington 18
- Slides: 18