AIMING FOR THE MOON Machine Learning Research for Team Astrobotics - - PowerPoint PPT Presentation

▶

Mar 17, 2023 292 likes •595 views

AIMING FOR THE MOON Machine Learning Research for Team Astrobotics Lunar Lander Research conducted at CMU in conjunction with Research Behind Lunar Lander Challenges of a Lunar Lander ML Research for the Lander - FTW Lander Perception /

SLIDE 1

AIMING FOR THE MOON

Machine Learning Research for Team Astrobotic’s Lunar Lander

Research conducted at CMU in conjunction with

SLIDE 2

Research Behind Lunar Lander

Challenges of a Lunar Lander ML Research for the Lander - FTW Lander Perception / Visual Registering Upcoming Work for Dec. 2012 Launch

SLIDE 3

Motivation of Lunar Exploits

Lunar Resources

 Unfiltered solar energy  Clean fusion (helium3)  Methane, ammonia, minerals

Lunar docking bay

 No gravity  Available fuel

Watchtower

 Base for infrared telescopes (icy craters)

SLIDE 4

Research Behind Lunar Lander

Challenges of a Lunar Lander ML Research for the Lander - FTW Lander Perception / Visual Registering Upcoming Work for Dec. 2012 Launch

SLIDE 5

It has been done

1969 – Apollo 11

 manned mission

1970 – Luna 17 with Луноход (Lunokhod 1)

 remote controlled robot

View from Camera 2 Images from Wikipedia and www.mentallandscape.com

SLIDE 6

So where is the novelty?

Pinpoint Landing - 500 m

 Less than a hundredth of a degree (lat/long)

High autonomy

 Little or no human support in landing  Preset trajectory  Fault detection  Error recovery

Proof of concept – reliable commercial lander

SLIDE 7

Stages of Landing

Orbit insertion

 Determine Lander orientation

SLIDE 8

Stages of Landing

Orbit insertion

 Determine Lander orientation

De-orbit and breaking

 Determine orbital parameters

SLIDE 9

Stages of Landing

Orbit insertion

 Determine Lander orientation

De-orbit and breaking

 Determine orbital parameters

Descent – 18km to 500m

 Keep track of Lander coordinates

SLIDE 10

Stages of Landing

Orbit insertion

 Determine Lander orientation

De-orbit and breaking

 Determine orbital parameters

Descent – 18km to 500m

 Keep track of Lander coordinates

Touch down

 500m to 150m: compute slopes, detect craters  Less than 150m to surface: detect small obstacles

SLIDE 11

Challenges – Perception

Detect whether Lander is off course
Detect whether sensors function properly
NO CAMERA - for a while
In case of off-course landing, pick landing spot

SLIDE 12

Research Behind Lunar Lander

ML Research for the Lander - FTW Lander Perception / Visual Registering Upcoming Work for Dec. 2012 Launch

SLIDE 13

Research - Vision

Landscape identification
Feature tracking

 Mega-structures  Craters

Obstacle detection

SLIDE 14

Research – Evidence Fusion

Density Elevation Map (DEM) construction

 Sparse LIDAR readings  Images of surface

Combine sensor readings to obtain position
Voting scheme to determine faulty

components

SLIDE 15

Research – Knowledge and AI

Decision unit in the Lander
Inference to determine

 Position  Lander State

Planning

 Save fuel resources  Pick landing spot to facilitate rover movement

SLIDE 16

Research Behind Lunar Lander

Lander Perception / Visual Registering Upcoming Work for Dec. 2012 Launch

SLIDE 17

Lander Sensor Array

Stage Requirement Sensor Orbit insertion Attitude Sun position Orbital Parameters IMU Sun tracker/Star tracker Pre-defined Deorbit Attitude Camera/IMU Descent stage (18kms to 500mts) Attitude Planning Camera/IMU Touch down (500mts to ground) Attitude Altitude Slope of Ground Velocity Surface characteristics Camera/IMU Pointed RADAR LIDAR Doppler LIDAR/Camera

SLIDE 18

Visual Registering

Step 1: Crater Detection

Crater Detection on LRO Image Crater Detection on Image captured by camera on the Black Magic platform

SLIDE 19

Test Set True Positives False Positives True Negatives False Negatives Apollo 11 4/5 3/20 17/20 1/5 Apollo 14 3/5 4/20 16/20 2/5 Apollo 16 5/5 3/20 17/20 0/5 Apollo 17 4/5 4/20 16/20 1/5

Step 2: Comparing Landscape to Stored Data

Visual Registering

SLIDE 20

Digital Elevation Map

Sparse LIDAR data (Lunar Reconnaissance Orbiter

SLIDE 21

Digital Elevation Map

Image (LCROSS Impact)

SLIDE 22

Markov Random Field

Image of cabeus crater LIDAR (1% of available data) X Z X – image Z – sparse elevation measurements Y – estimated elevation map L – points where elevation readings exist N(i) – neighborhood of point i on the grid wij – correlation between pixels in image

SLIDE 23

MRF with Shading Coefficient

X – image Z – sparse elevation measurements Y – estimated elevation map L – points where elevation readings exist N(i) – neighborhood of point i on the grid wij – correlation between pixels in image Pixel correlation Shading coefficient

p =5% quantile of image data – the shaded pixels

Overlap of image and LIDAR

SLIDE 24

Experiment - Terrain Model

SLIDE 25

Experiment – Model + LIDAR

SLIDE 26

MRF Results

Elevation map after 200 iterations of Coordinate Descent

mean error 143.75 m
13.2% of average elevation

Elevation map after interpolation

mean error 175.20 m
16.09% of average elevation

SLIDE 27

Scanning for a landing site

Image-based landing site selection
Elevation-based landing site selection

SLIDE 28

Research Behind Lunar Lander

Upcoming Work for Dec. 2012 Launch

SLIDE 29

Work for December 2012 Launch

Feature Tracking
AI unit

 Decisions  Inference  Planning

Integrated Testing Framework