The Caltech Multi-Vehicle Wireless Testbed: Initial Implementation

Transcription

1 The Caltech Multi-Vehicle Wireless Testbed: Initial Implementation Lars Cremean, Jason Meltzer, Ben Lee, Richard Murray (In collaboration with Steven Low, Jason Hickey, Eric Klavins, Reza Olfati, Bill Dunbar and Dave van Gogh) Nov Conference on Cooperative Control and Optimization 1

2 Outline What is the Multi-Vehicle Wireless Testbed (MVWT)? Implementation of closed-loop vehicle control Preliminary control results Extension to multiple vehicles Conclusions and future work Nov Conference on Cooperative Control and Optimization 2

3 What is the MVWT? Experimental platform for investigating the intersection of control, computation and communications Consists of: A number of wireless vehicles Vision system (emulates GPS) Command system Inter-vehicle communication Nov Conference on Cooperative Control and Optimization 3

4 Prototype Vehicle Design Actuation: Two high-power unidirectional ducted fans No onboard sensing Receive fan input signals via radio link Nov Conference on Cooperative Control and Optimization 4

5 Video Manual Control Nov Conference on Cooperative Control and Optimization 5

6 Vehicle Dynamics Assuming linear friction acting at CoM vehicle is linear in state, nonlinear in input Input-constrained, underactuated (0 F S/P F max ) Nov Conference on Cooperative Control and Optimization 6

7 Current Physical Layout Vision PC Command & Control PC dspace card Programmable Interrupt Controller (PIC) Trainer Port Cable RC Transmitter Nov Conference on Cooperative Control and Optimization 7

8 Future Physical Layout Vision PC Command PC dspace card (optional) Vision computer broadcasts states of all vehicles via wireless ethernet (UDP) Command PC broadcasts individual or coordinated tasks Each vehicle has own control and communication Nov Conference on Cooperative Control and Optimization 8

9 Next Generation Vehicle Design Smaller, lighter Remove radio control dependency On-board laptop processing for each vehicle Receive position data via wireless CAD courtesy of Kelly Klima Nov Conference on Cooperative Control and Optimization 9

10 Project Structure Vehicle(s) Command Center Simulation Planning Planning Group M-ship NTG User Interface Group M-ship NTG Vehicle Control Vehicle Control Real-Time OS Real-Time Monitor HW Emulator Onboard Comp. Onboard Comms. LPS Comms (NS) Vehicle HW Vision System Physics (WM2D) Nov Conference on Cooperative Control and Optimization 10

11 Vision System Four monochrome 648x484 CCD cameras mounted on ceiling Four Matrox Genesis vision processing boards C++ code for processing, calibration and data broadcast via UDP Vehicles identified via blob analysis Calibration: J.Y. Bouguet s Calibration Toolbox, (Caltech vision lab) Nov Conference on Cooperative Control and Optimization 11

12 Camera Image Nov Conference on Cooperative Control and Optimization 12

13 Vision Calibration Nov Conference on Cooperative Control and Optimization 13

14 Vision Performance Specs 60 Hz 16.8mm max. calibration error 90% 8mm ~16msec latency Nov Conference on Cooperative Control and Optimization 14

15 Simulation Environment Dynamic simulation by Working Model 2D DDE interface to MATLAB (and C/C++) for control design Collision detection/ handling Nov Conference on Cooperative Control and Optimization 15

16 Initial Controller Design Two PID loops based on deconstructed state Inner loop preserves angle to target Outer loop attempts to regulate distance to target to zero Gain applied to outer loop depends on inner loop performance F K ref d P x,y T K θ Nov Conference on Cooperative Control and Optimization 16

17 Initial data Theta control Nov Conference on Cooperative Control and Optimization 17

18 Movie Point-to-point automatic control Nov Conference on Cooperative Control and Optimization 18

19 Data Point-to-point automatic control Nov Conference on Cooperative Control and Optimization 19

20 Trajectory Tracking - Methodology More general framework for control Compute feasible trajectory in real-time Compute associated inputs Control about nominal trajectory D ref Reference: Milam, Mushambi, Murray, A Computational Approach to Real-Time Trajectory Generation for Constrained Mechanical Systems, CDC 2000 Trajectory Generation u d x d noise δ u Plant P Local Control output Nov Conference on Cooperative Control and Optimization 20

21 Extension to Multiple Vehicles Currently two prototype vehicles built Eight next-generation vehicles planned Each vehicle, in general, will have knowledge of the states of itself and some subset of all other vehicles Nov Conference on Cooperative Control and Optimization 21

22 Preliminary Graph Theory Results 5 6 Model information topology as a directed graph Changes in interconnection affect stability of configuration Nyquist-based criterion for formation stability Reference: Fax, Murray; Information Flow and Cooperative Control of Vehicle Formations, IFAC 2002 submitted Nov Conference on Cooperative Control and Optimization 22

23 Inter-vehicle Communication Protocols for inter-vehicle communication in development Potential for use of both and Bluetooth wireless communication Graph theoretic aspects extend to communicated information as well as sensed information Nov Conference on Cooperative Control and Optimization 23

24 Benchmark Multi-Vehicle Tasks Split/rejoin maneuvers Formation keeping while tracking a common trajectory Formation changing Nov Conference on Cooperative Control and Optimization 24

25 Conclusions Have implemented reliable closed-loop control infrastructure for coordinated control of multiple vehicles Nov Conference on Cooperative Control and Optimization 25

26 and Future Work advanced controller design next generation vehicle design trajectory generation/following multiple vehicle implementation for limited sensing network network protocol design for inter-vehicle communication testing new algorithms/strategies/theory Nov Conference on Cooperative Control and Optimization 26