Intelligent Instruments. Kevin H. Knuth Department of Physics University at Albany

Transcription

1 Intelligent Instruments Kevin H. Knuth Department of Physics University at Albany

2 Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information? From "The Rock" by T.S. Elliot

3 Remote Science Opportunity on Mars More and more are our instruments required to perform science operations further from the intervention of humans. Dust devils whip across Gusev Crater on Mars

4 Intelligent Autonomous Instruments Require: Stability Control Instrument Health Monitoring Automated Calibration Accurate Onboard Data Analysis Adequate Data Coverage Ability to Actively Seek Data

5 Novel Instrument Design To accomplish these goals, these novel instruments must Monitor their own state (health and calibration) Infer their state from self-sensing Be equipped with dense sensor networks Infer calibration parameters Learn from data Make inferences from data Perform hypothesis testing Ask new questions Actively seek new data Select optimal experiments

6 Sound Familiar? Your frontal lobes carry a model of yourself that is continually updated from data received from a dense sensor network. This implements both Instrument Health Monitoring and Calibration You learn from new data by updating your model of the world. You actively seek new data by asking relevant questions.

7 Body and Brain form a Symbiotic Unit

8 Instruments and Data Analysis are Disjoint

9 The Basic Components

10 Intelligent Instrument Test Bed

11 NASA Funded Research Intelligent Symbiotic Instrument Analysis Systems

12 The Dispersed Fourier Transform Spectrometer Courtesy: Dr. Arsen Hajian

13 The Alignment Schematic Light (blue) enters the spectrometer through a fiber optic launcher. A laser beam (red) is injected into the system for metrology. It is filtered out of the results with optical notch filters. Cameras C1-8 collect data for the autoalignment system. Barriers B1-6 can be introduced via autoalignment system to test various sub-systems.

14 Auto-Alignment and Stabilization We can align dfts from anywhere in the world Alignment required 5 minutes on Code in Java (cross-platform) Modified our full-aperture metrology system to create an active compensation system Compute f(t) with respect to f(t o ) using lock-in-amp (SR830) Feed error signal back to delay line (Parker) Refine results with further Bayesian calibration

15 Problems and Solutions Problem: Collaborator afraid we would steal the design of his instrument. Solution: Design and Construct our own Instrument Problem 1: Collaborator concerned about having automated software run his expensive instrument. Solution: Instrument must be inexpensive in the event of a catastrophe Problem 2: NASA management confused instrument already takes data in an automated fashion. Solution 1: New management Solution 2: Instrument must be OBVIOUSLY Intelligent. Problem: NASA cuts funding to entire program Solution: Instrument must be inexpensive Solution: Secure funding from alternate sources

16 The LEGO Mindstorms NXT System 1 The NXT Brick is the brain of the system. 2 Touch Sensor 3 Microphone 4 Light Sensor 5 Ultrasonic Rangefinder 6 Servo Motors

17 Lego teams with HiTecnic NEW! Accelerometer Prototype Board Color Sensor Digital Compass Sensor and Motor Multiplexers

18 NXT Communicates with Laptop Software Hardware SYMBIOTIC!

19 LDRAW (Lego Cad System)

20 Rendering and Animating Designs Created by Kevin Knuth 2007 with LDraw and POVRay

21 Advantages of Lego NXT 1. Cost: One NXT Robotics Kit = $ Design: Robot Bodies are not constrained 3. Construction: Robots can be built in a matter of hours 4. Documentation: LDraw software allows one to thoroughly document the robot s construction. One can generate Parts Lists, Keep design or send to many others. 5. Brick: The Brick is programmable, but can also be set up to talk to a computer. This enables complex software to effectively run in reasonable times for Real Time adventures.

22

23 Characterization with a Light Sensor GOAL: Characterize the circle: {x, y, r} with as few measurements as possible CONSTRAINT: Only point measurements are allowed

24 Sample from the Posterior After several measurements, the posterior in {x, y, r}-space becomes well-localized. Here are possible solutions sampled from the posterior using nested sampling

25 Sample from the Posterior By querying each sampled circle, we can obtain a set of hypothesized measurements for each possible measurement location. From this we create an entropy map, which tells us where we can expect to obtain the greatest amount of information.

26 NXT Steps

27 Measure Acoustic Radiation Pattern

28 Laser Scanner

29 Which Experiment? In the future, he won t need help from humans to determine the best experiment to perform. He ll decide for himself.

30 Acknowledgements: NASA SISM IS Program (Knuth) SIM Preparatory Science Program (NRA 98-OSS-07) (Hajian) Thanks Arsen Hajian USNO J. Pat Castle EA/NASA Ames Nikolay Lvov QSS/NASA Ames John Stutz NASA Ames Dogan Timucin NASA Ames Kevin Wheeler NASA Ames Brian Pohl UNC Chapel Hill J. Thomas Armstrong NRL David Mozurkewich NRL Robert B. Hindsley NRL Christopher Tycner Univ. of Toronto Robert Olling USNO P.S. There is nothing wrong with human intervention

31