3 rd Off Earth Mining Forum AUTONOMOUS SPACECRAFT NAVIGATION NEAR AN - - PowerPoint PPT Presentation

Slide 0

3rd Off Earth Mining Forum AUTONOMOUS SPACECRAFT NAVIGATION NEAR AN ASTEROID Arunkumar Rathinam, PhD candidate, ACSER, UNSW Sydney

Slide 1

Navigation near an asteroid

Major factors – size – irregular shape – weak gravitational force – non-gravitational perturbations Navigation in previous asteroid exploration missions – radiometric tracking (two-way Doppler, two-way range, Delta-DOR) – in combination with on-board optical data (based on landmark locations)

Slide 2

Future Deep Space Missions

list of Mars missions launching in 2020: – NASA’s Mars 2020 rover, ESA’s ExoMars 2020 rover, China’s orbiter/lander, UAE’s Hope orbiter, India’s Mars Orbiter Mission-2, SpaceX - Red Dragon Mars lander Other missions possibly using DSN in 2020: – NASA’s Odyssey orbiter, MAVEN orbiter and MRO, ESA’s Mars Express orbiter and TGO, India’s MOM, New Horizon, Voyager, OSIRIS- REx, Hayabusa-2

Spacecraft autonomy is a priority

http://spacenews.com/mars-looming-traffic-jam/

Slide 3

Mission design

• Asteroid characterisation phase (3-6 months)

– by global mapping and observation – hovering around a home position (10~20 km above surface) – Position determinations based on radio, star-based nav. Techniques » Multiple descent operations - to determine the asteroid gravity by LIDAR and two-way Doppler measurements – Estimate unknown parameters the spin-axis orientation, rotation period – Generate shape model of the asteroid, with a core set of surface landmarks – Transition from star-based to landmark-based optical navigation - next phase

Slide 4

Shape model for Navigation

• From the images of asteroid

– generate Maplet - small scale 3D high resolution maps – Stereophotoclinometry – Each maplet was centered on a landmark

• Shape reconstruction

– Limb profile to generate geometric shape – assemble maplets on reconstructed shape

• landmark table – helps identification and tracking of landmarks

Shape model of asteroid Itokawa – Hayabusa mission

Slide 5

Challenges in Autonomous Navigation (1)

• Navigation and mapping - mutually dependent problems
• Optical navigation

– poor illumination of asteroid surface – Stereo cameras and laser range finders won’t work at higher altitude (~20km) – need for robust image processing and data association

• Dynamics

– lack of accurate a priori knowledge of dynamic parameters – require good dynamic model » asteroid’s rigid body dynamics » Spacecraft’s motion (hover home pos. / establish stable orbit / manoeuvre control)

Slide 6

Challenges in Autonomous Navigation (2)

• Close proximity navigation

– High accuracy demands – major perturbations » non-gravitational forces from solar radiation pressure » pressure exerted by re-emitted IR radiation from the spacecraft and the asteroid

• Controlled manoeuvre

– small delta-V (e.g. TAG, delta-V’s range between 1 ~ 20 cm/sec)

• To reduce uncertainties

– frequent orbit determinations and ephemeris updates – BUT, adds burden on the navigation teams

Slide 7

Simultaneous Localization and Mapping (SLAM)

• Estimate the robot’s pose and the map of the environment at the same time
• No need for any a priori knowledge of environment
• SLAM is chicken-egg problem

– map is needed for localization – pose estimate is needed for mapping

• In probabilistic form : , | :, :,

– observation model | , – motion model | ,

[Durrant‐Whyte and Bailey ‘06 ]

Slide 8

SLAM Framework

Map Estimation feature extraction

Data association

• short-term (feature matching)

Long-term (loop closure)

front-end back-end Sensors

integrate multiple sensor data Images – ONC, Attitude - star sensors, Position - radiometric ranging, Inertial - maneuver using thrusters, LIDAR/Laser altimeter, Point cloud data from Flash LIDAR

estimate

Slide 9

SLAM approach

• Filtering approach

– EKF SLAM approach » robot pose and the environment feature positions in one state vector » quadratic nature of the map with increasing number of landmarks – Particle filter » maintains multiple map hypotheses, each conditioned on a stochastically sampled trajectory through the environment. » computationally intensive

• Optimisation approach

– Graph SLAM – states are represented as graph nodes

Slide 10

SLAM - Factor graph

– Factor graph is a bipartite graph with two types of node: variable node and factor node – Variable nodes (blue) constitutes the estimated state (xk and li) – Factor nodes (green) represents the joint probability distribution between the states – f , Gaussian probability distribution b/w random variables a & b; error between the variables must be minimized – Factors encode all the information entering the system – Graph captures the way this information is propagated to the hidden states

Slide 11

Graph SLAM for asteroid navigation

Factor graph representation of variables (spacecraft and asteroid state, landmark position), factors (measurement and motion)

Slide 12

Preliminary experiment

Experiment setup in MATLAB – asteroid’s diameter approx. 500 m – with 200 distinct features randomly distributed landmarks – rotational period : 8hr – the spacecraft orbiting at 600 m from the center – entire simulation covers a duration of about 120 key frames – revisit the landmarks between 2∼3 times

Simulated Asteroid model with landmarks

Slide 13

Summary

– Future work » Estimate the other dynamics parameters in the process of state estimate » Develop the framework with integrating other sensor data » Want to achieve long term robustness

Reconstructed model with landmarks and spacecraft positions

Slide 14

THANKS

Slide 15

de Santayana, R. Pardo, and M. Lauer. "Optical measurements for rosetta navigation near the comet." Proceedings of the 25th International Symposium on Space Flight Dynamics (ISSFD), Munich. 2015. Cadena, Cesar, et al. "Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age." IEEE Transactions on Robotics 32.6 (2016): 1309- 1332.

Slide 16

Motion model

• Asteroid’s dynamics model

X =

• q 1

2 W ω J J ω Q

• ω
• Spacecraft motion model

=

• q 1

2 W Q ω J J Q

Slide 17

, ,

• , ,
• , ∝
• ,
• ,

, ∝

• ,
• ,

, , , ,

• Factor graph - formulation

Robot motion Landmark measurements Factors Errors Robot motion Landmark measurements