Robotics in Planetary Exploration

1
Robotics in Planetary Exploration

• Lorenza Saitta
• Attilio Giordana
• Ugo Galassi
• Dipartimento di Informatica
• Università del Piemonte Orientale
• saitta,giordana,galassi_at_mfn.unipmn.it

2
Outline

• Autonomous planetary exploration
• Landing
• Craters and hazard detection
• Rover navigation
• Conclusions

3
Autonomous planetary exploration

• Increasing the level of spacecrafts or robots
  autonomy is essential for planetary exploration
• Main problems
• Communication latency
• Bandwidth limitations
• Computer Vision can play an important role in
  developing (semi-)autonomous systems for future
  Mars exploration
• The focus is on two mission phases
• the ?nal descent of the lander
• the environment exploration by the part of a rover

4
Landing

• Landing sites for space exploration missions have
  historically been determined off-line
• Landing sites with high scienti?c potential are
  not considered if safety cannot be guaranteed
• Future missions require the ability to
  autonomously perform the critical landing
  operations
• Hazard detection
• Retargeting to a new landing site
• Pinpoint landing will be required
• Missions targeted at specific types of surface or
  terrains
• Landing near other prepositioned surface assets

5
Vision-powered landing

• Navigation accuracy during EDL increased using
  camera measurements
• Cameras input can be used for identifying
• Terrain features (slope, roughness)
• Landmarks
• Craters are very commonly used landmarks.
• They are abundant
• Their detection can be carried out ef?ciently,
  under varying image scale, viewpoint, and
  illumination conditions.

6
Landing process

• Small lander scenario applied to Mars
• Delivery of 1.5 tons on the Mars surface
• Precise, safe soft landing
• Complete autonomy

• Atmospheric phase
• start 120?125 km above surface
• speed 5500 to 600 m/s
• duration 250 s

• Powered descent
• start 4 km above surface
• speed 50 to 0 m/s
• duration 120 s

• Parachute phase
• start 10 km above surface
• speed 600 to 50 m/s
• duration 180 s

7
Craters detection
Original
Filtered
Bright Threshold Dark
Threshold
Binarization
Thresholding Final image
8
ML for craters detection

• Machine learning approaches can be used for
  craters detection.
• Classical machine learning approaches (neural
  networks, decision trees and support vector
  machines)
• More sophisticated approaches can be used
  (boosting)

9
Hazard voidance

• Hazard map components are
• Craters
• Steep slopes
• Discontinuities
• Shadows
• It is required a detection rate of 99, with
  false alarm rate of 1.

Hazard level
10
Rover

• It is limited by
• the daily available energy
• the mission duration
• Autonomous local obstacle avoidance and on- board
  global planning strategy systems are required to
  maintain an important daily traverse rate.
• The robot build up a map of the environment,
  using a stereocamera.

11
Navigation map
Dangerous regions Forbidden region
Navigable region Uncertain navigable
region Unknown region
12
Machine learning and planetary exploration

• The following methodologies have been designed on
  purpose for path planning and trajectory control
• Scope learning
• Hybrid learning
• Human mimicking

13
Conclusions

• I have described some problems, arising in
  planetary exploration, that o?er a great
  potential for application of Machine Learning
  techniques.
• The focus has been set on tasks that strongly
  rely on visual input.
• The expected results are su?ciently general to
  transcend the ?eld of planetary exploration, and
  can be used also on Earth in various industrial
  environments