AI in Space Exploration

1
AI in Space Exploration

• Stephen Dabideen
• Yizenia Mora

2
Agenda

• Planning and Scheduling (CASPER)
• Autonomous Navigation (AutoNav)
• Communications with Earth (Beacon)
• Autonomous Onboard Science (ASE OASIS)
• Data Mining (SKICAT)

3
Autonomous Navigation (AutoNav)

• What is AutoNav?
• Autonomous Optical Navigation system uses an
  expert-system-like architecture to guide a
  spacecraft to its target, first used in DS1
• Enables a spacecraft to navigate independently of
  ground teams and ground links
• It commands the ion propulsion system and the
  spacecraft's altitude control system to change
  trajectory as needed
• AutoNav also determines how much power to devote
  to the ion propulsion system
• Use location to determine how much energy
  generated by solar array
• Intended to be reusable

4
Autonomous Navigation (AutoNav)

• Subsystems functions
• Navigation executive function
• Controls all AutoNav operations that cause
  physical action by spacecraft.
• Optimizes time utilization by planning turn
  sequences
• Image processing
• Integrates camera and imaging spectrometer to
  take pictures of asteroids and stars, to
  determine its location
• 0.1 pixel accuracy
• Orbit determination
• Uses a batch-sequential modified Kalman filter to
  compute the spacecrafts position
• Maneuver planning
• Use OD to compute updates to upcoming trust plan.

5
Communications with Earth (Beacon)

• Spacecraft determines when ground support is
  needed and what information is relevant
• Advantages
• Reduces costs of the spacecraft-to-ground link
• Downlinks only pertinent information

6
Communications with Earth (Beacon)

• Two subsystems
• Subsystem 1
• End-to-end tone system to inform the ground
  whether data needs to be sent
• One of four possible requests (no action
  required, contact when convenient, contact within
  a certain time, or contact immediately)
• Subsystem 2
• Produce intelligent data summaries to be
  downlinked as telemetry when ground responds to
  tone request
• Four types of engineering telemetry
• High-level spacecraft information since the last
  ground contact
• Episode data
• Snapshot telemetry
• Performance data
• ELMER used to detect anomalies

7
Communications with Earth (Beacon)Detecting
Anomalies (ELMER)

• Traditional thresholds
• Static, manually predefined red lines
• A lot of false alarms
• ELMER (Envelope Learning and Monitoring using
  Error Relaxation)
• Time-varying alarm thresholds
• Neural networks
• Trained with nominal sensor data
• High- and low-expectation bounds (envelopes)

8
Autonomous On-Board Science

• The dream
• An autonomous Mars rover traversing the planets
  surface for a couple of years, unattended by
  humans, collecting and catching samples

9
Autonomous On-Board Science

• The dream
• An autonomous Mars rover traversing the planets
  surface for a couple of years, unattended by
  humans, collecting and catching samples
• Reality check Spirit and Opportunity
• 4 drivers per rover
• About 20 simulations per move
• Remote-controlled over 150 million miles away
• Opportunity's farthest distance to date 15 m

10
Autonomous On-Board Science

• Need for automated science
• Slim window of opportunity for discovery
• Autonomy can provide more reactive, flexible
  architecture to respond to unanticipated events
• Limited downlink bandwidth
• Time delay
• Accomplishments thus far
• New method analyzing visible broadband images
  using neural networks
• Important features extracted and combined with
  spectral classifications and decisions are made
  using a decision tree directed towards specific
  goals
• Analysis of spectral data
• Hierarchy of neural nets place spectra into
  progressively more detailed geologic classes
• Decompose mixtures from unknown spectra
• Will help automate characterization of planetary
  surface

11
Autonomous On-Board Science (ASE)

• The Autonomous Sciencecraft Experiment (ASE)
• Used on Earth Observing One (EO -1)
• Demonstrates integrated autonomous science
• Features several science algorithms including
• Event detection
• Feature detection
• Change detection
• Analyzes to detect trigger conditions such as
  science events
• Based on these observations CASPER will replan
• Science analysis techniques include
• Thermal anomaly detection
• Cloud detection
• Flood scene classification

12
Autonomous On-Board Science (OASIS)

• Onboard Autonomous Science investigation System
• Due to limited bandwidth, rovers must
  intelligently select what data to transmit
  back to Earth
• How?
• Machine leaning techniques to prioritize data
• The capability of OASIS enables a rover to
  perform data collections which were not
  originally planned, even without having to wait
  for a command from Earth
• Researchers are interested in
• Pre-specified signals of scientific interest
• Unexpected or anomalous features
• Typical characteristics of a region
• OASIS has different levels of autonomy, from
  following a predefined path and taking only
  planned measurements to commanding the rover to
  deviate slightly from path to get new measurements

13
Data Mining (SKICAT)

• What?
• SKy-Image Cataloging and Analysis Tool
• Assign galaxies and stars to known classes and
  identify new classes
• Why?
• Databases are too large for an astronomer to
  analyze manually
• How?
• Automated Bayesian classification
• Attributes such as brightness, area, color,
  morphology,
• Training data consisting of astronomer-classified
  sky objects
• Classifiers applied to new survey images

14
Data Mining (SKICAT)

• Results
• One of the most outstanding successes
• 1,000-10,000 times faster than astronomers
• More consistent classification
• Able to classify extremely faint objects
• Astronomers freed for more challenging analysis
  and interpretation
• Comprehensive catalog of approximately 3 billion
  entries