Year: 2025
Type: conference-paper
Abstract: Trajectory optimization is a cornerstone of modern robot autonomy, enabling systems to compute trajectories and controls in real-time while respecting safety and physical constraints. However, it has seen limited usage in spaceflight applications due to its heavy computational demands that exceed the capability of most flight computers. In this work, we provide results on the first in-space demonstration of using machine learning-based warm starts for accelerating trajectory optimization for the Astrobee free-flying robot onboard the International Space Station (ISS). We formulate a data-driven optimal control approach that trains a neural network to learn the structure of the trajectory generation problem being solved using sequential convex programming (SCP). Onboard, this trained neural network predicts solutions for the trajectory generation problem and relies on using the SCP solver to enforce safety constraints for the system. Our trained network reduces the number of solver iterations required for convergence in cases including rotational dynamics by $60 \%$ and in cases with obstacles drawn from the training distribution of the warm start model by $50 \%$. This work represents a significant milestone in the use of learning-based control for spaceflight applications and a stepping stone for future advances in the use of machine learning for autonomous guidance, navigation, & control.
Institutions: Stanford University, Johns Hopkins University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Awards: THE GOAL OF THIS RESEARCH IS TO INTEGRATE ADHESION-BASED GRASPING WITH CONTROL AND PLANNING TO ENABLE ASSISTIVE FREE-FLYERS TO GRASP AND MANIPULATE PAYLOADS SUPPLIES AND TOOLS INSIDE AND OUTSIDE SPACECRAFT. THE TECHNICAL OBJECTIVES ARE: 1. DESIGN ANALYZE AND TEST MECHANISMS FOR GRASPING AND MANIPULATION THAT EMPLOY CONTROLLABLE ADHESION AND GRASP AND MANIPULATE PAYLOADS WITHIN SPACE VEHICLES AND ULTIMATELY IN EXTRAVEHICULAR OPERATIONS. 2. DEVISE CONTROL AND PLANNING ALGORITHMS THAT ALLOW FREE-FLYERS TO NAVIGATE ACROSS COMPLEX SPACE STRUCTURES AND GRASP AND MANIPULATE PAYLOADS WITH ADHESION-BASED APPENDAGES. 3. VALIDATE THE TECHNOLOGY FOR MANIPULATION TASKS ON A STATE-OF-THE-ART TEST BED AT STANFORD EMULATING THE MOTION OF SPHERES AND FUTURE ASTROBEES IN ZERO GRAVITY. OUR TECHNICAL APPROACH IS BASED ON GECKO-INSPIRED CONTROLLABLE ADHESIVES THAT ARE NON-STICKY IN THE DEFAULT STATE AND ADHERE ONLY WHEN LOADED IN A CERTAIN DIRECTION. RESEARCH FROM THE INVESTIGATORS HAS DEMONSTRATED THAT SUCH ADHESIVES SCALE TO LARGE LOADS FUNCTION IN VACUUM AND OVER A WIDE RANGE OF TEMPERATURES AND PERFORM WELL UNDER COLLISIONS IN A VARIETY OF APPLICATIONS INCLUDING CLIMBING ROBOTS HUMAN CLIMBING AND PERCHING MICRO-AIR VEHICLES. IN THE PROPOSED EFFORT WE WILL DESIGN APPENDAGES WITH SPECIALIZED GRIPPERS AND WRISTS THAT LEVERAGE SUCH ADHESIVES. THE WRISTS WILL BECOME FREELY PIVOTING IF AN OVERLOAD TORQUE IS APPLIED (E.G. IN CASE OF A COLLISION). WRISTS AND GRIPPERS WITHOUT A FREE-FLYER COULD ALSO BE USED BY HUMANS WHO WANT TO ATTACH CAMERAS SENSORS OR OTHER DEVICES TO FLAT OR CURVED SURFACES. FREE-FLYERS WITH ADHESION WILL REQUIRE NOVEL ALGORITHMS FOR PLANNING AND CONTROL THAT TAKE INTO ACCOUNT THE COUPLED DYNAMICS OF THE FREE-FLYER AND THOSE OF THE END EFFECTOR AND ALLOW THE APPENDAGES TO RELIABLY ENGAGE OR RELEASE PAYLOADS. THE APPROACH WILL BE TO TAILOR RECENT BREAKTHROUGHS BY THE INVESTIGATORS IN SPACECRAFT MOTION PLANNING SPECIFICALLY SAMPLING-BASED KINODYNAMIC PLANNING WITH ASYMPTOTIC OPTIMALITY GUARANTEES. ADDITIONALLY WE WILL DESIGN ADAPTIVE MODEL PREDICTIVE CONTROL ALGORITHMS BASED ON CONVEX OPTIMIZATION THAT ALLOW PRECISE DOCKING DURING THE GRASPING AND RELEASE AND ARE ROBUST TO UNCERTAINTIES IN THE PAYLOAD'S MASS AND INERTIA. THE ALGORITHMS WILL EXPLOIT THE UNIQUE FEATURES OF CONTROLLABLE ADHESIVES WHICH HOLD THE POTENTIAL TO CONSIDERABLY EASE MANIPULATION TASKS AS COMPARED TO CONVENTIONAL GRIPPERS THAT USE SPECIALIZED GRAPPLING FEATURES. THIS PROJECT FOCUSES ON GRASPING AND MANIPULATION WITHIN SPACE VEHICLES BUT ALSO PAVES THE WAY FOR EXTRA-VEHICULAR OPERATIONS WHERE FEW ALTERNATIVE TECHNOLOGIES ARE EFFECTIVE. IT BRINGS TOGETHER A TEAM OF EXPERTS IN DEXTEROUS MANIPULATION GRASPING DEVICES MOTION PLANNING AND CONTROL ASTRONAUTICS AND HUMAN SPACEFLIGHT FROM STANFORD AND MIT. IT BUILDS ON NEW RESULTS THAT HAVE RECEIVED TWO BEST-PAPER AWARDS AT MAJOR ROBOTICS CONFERENCES. THIS PROPOSAL ALSO BUILDS ON A COLLABORATION WITH DR. AARON PARNESS AT JPL TO INVESTIGATE GECKO-INSPIRED ADHESIVES FOR MICRO-GRAVITY APPLICATIONS AND SPACE-READINESS. RESPONSIVENESS: THIS PROPOSAL RESPONDS TO TOPIC 1: PAYLOAD TECHNOLOGIES FOR ASSIS- TIVE FREE-FLYERS. IT ADDRESSES THE INTEREST FROM NASA TO DEVELOP ASSISTIVE FREE-FLYERS THAT CAN GRASP AND MANIPULATE PAYLOADS SUPPLIES AND TOOLS INSIDE AND OUTSIDE SPACE VEHICLES. EXAMPLE TASKS INCLUDE UNPACKING AND STORING SUPPLIES POSITIONING SENSORS AND CAMERAS AND FETCHING TOOLS AND OTHER ITEMS FOR ASTRONAUTS. SPECIFICALLY IT ADDRESSES TECHNOLOGIES HIGHLIGHTED IN THE 2015 NASA SPACE TECHNOLOGY ROADMAPS REPORT INCLUDING: TERRAIN ADHESION (#4.2.8.3) GRAPPLING (#4.3.7.1) FREE-FLOATING ROBOTS (#4.2.4.1) AND ONBOARD TRAJECTORY PLANNING AND OPTIMIZATION ALGORITHMS (#5.4.2.3)., CUBESATS ARE A UBIQUITOUS SATELLITE PLATFORM COMPOSED OF 10X10X10CM "1U" UNITS WITH HUNDREDS OF MISSIONS FLOWN. CUBESATS INCREASINGLY REQUIRE 3-AXIS CONTROL FOR POINTING THE SCIENCE PAYLOAD OR COMMUNICATIONS ANTENNA AND DUE TO MASS AND VOLUME CONSTRAINTS USE A SUITE OF REACTION WHEELS FOR IMPLEMENTING FINE CONTROL. THROUGH THE COURSE OF AN ORBIT EXTERNAL DISTURBANCE TORQUES ACTING ON THE CUBESAT REACTION WHEEL SYSTEM SATURATE AND RESULT IN A LOSS OF CONTROL FOR THE CUBESAT. IN THIS CASE MAGNETIC TORQUERS ACTUATE MOMENTUM DUMPING BY PROVIDING A TORQUE TO DESATURATE THE SYSTEM. RATHER THAN USING MAGNETIC TORQUERS I PROPOSE A METHODOLOGY USING AERODYNAMIC TORQUES TO DESATURATE REACTION WHEELS. AS THE SYSTEM STARTS TO SLOWLY LOSE CONTROL CAPABILITY DUE TO SATURATION THE REACTION WHEELS CAN BE USED PERIODICALLY TO ORIENT THE SPACECRAFT IN AN ORIENTATION WHERE THE RESULTING AERODYNAMIC TORQUE DESATURATES THE SYSTEM. AERODYNAMIC TORQUES ON A SPACECRAFT RESULT FROM DRAG IN LOW-EARTH ORBIT (LEO) ACTING ON THE SPACECRAFT MOMENT ARM. USING A METHODOLOGY DESCRIBED BY KARAMI AND SASSANI THE CUBESAT CAN BE RE-ORIENTED TO ALIGN THE MOMENT ARM IN A CONFIGURATION SUCH THAT THE AERODYNAMIC TORQUE DESATURATES THE SYSTEM. IN ORDER TO DETERMINE THIS ORIENTATION AN ACCURATE ESTIMATE OF THE CENTER OF PRESSURE MUST BE CALCULATED. THIS CENTER OF PRESSURE DETERMINES THE DIRECTION OF THE MOMENT ARM AND HENCE THE RESULTING AERODYNAMIC TORQUE DIRECTION. A MULTIPLICATIVE EXTENDED KALMAN FILTER (MEKF) ALGORITHM HAS BEEN PROPOSED THAT USES EULER'S RIGID-BODY EQUATIONS AND SENSOR MEASUREMENTS TO PROVIDE AN ACCURATE ESTIMATE OF THE LOCATION OF THIS CENTER OF PRESSURE. AFTER THE MEKF HAS BEEN RUN THE UPDATED CENTER OF PRESSURE ESTIMATE IS THEN USED TO IMPLEMENT THE KARAMI/SASSANI METHODOLOGY TO DESATURATE THE SYSTEM. THROUGH A COMBINATION OF THEORETICAL AND SIMULATION-BASED VERIFICATION METHODS THE GLOBAL STABILITY AND EFFICACY OF THIS AERODYNAMIC TORQUE DESATURATION METHOD CAN BE INVESTIGATED. CUBESATS TYPICALLY ALLOCATE 10-20 PERCENT OF VOLUME ON A 1U PLATFORM FOR MAGNETIC TORQUERS USED TO DESATURATE REACTION WHEELS. THE PROPOSED AERODYNAMIC TORQUE DESATURATION METHODOLOGY WOULD ALLOW FOR MORE VOLUME MASS AND POWER TO BE DEDICATED FOR AN ADDITIONAL RW EXPANDED SCIENCE PAYLOAD OR MINIATURE ELECTRIC PROPULSION THRUSTER TO ENABLE A GREATER CAPABILITY FOR MISSIONS. FURTHERMORE THIS METHODOLOGY TO USE EXTERNAL DISTURBANCE TORQUES TO DESATURATE REACTION WHEELS CAN BE EXTENDED TO CUBESAT MISSIONS FLYING TO PLANETARY BODIES WHERE MASS AND VOLUME CONSTRAINTS AND THE LACK OF A PLANETARY MAGNETIC FIELD RULE OUT THE USE OF THRUSTERS OR MAGNETIC TORQUERS FOR DESATURATION. THUS DOZENS OF CUBESAT MISSIONS IN ACADEMIA AND INDUSTRY THAT USE REACTION WHEELS COULD BENEFIT FROM THIS IMPLEMENTATION.
OpenAlex ID: W7140081306