TO ENABLE ROBUST AND RESILIENT SPACE-BASED ASSETS THAT ARE ABLE TO OPERATE AUTONOMOUSLY IN CHAOTIC GRAVITATIONAL ENVIRONMENTS NEW TECHNOLOGIES MUST BE DEVELOPED TO SUPPORT REGULAR PLANNING AND SCHEDULING OF EFFICIENT YET RELIABLE MANEUVERS. REDUCING THE DEPENDENCY ON A HUMAN-IN-THE-LOOP AND SHIFTING THESE RESPONSIBILITIES TO AN ONBOARD COMPUTING PLATFORM SUPPORTS RAPID IMPLEMENTATION OF MANEUVERS IN RESPONSE TO AN UNCERTAIN ENVIRONMENT. EVEN WITH CONTINUED IMPROVEMENTS IN COMPUTING EQUIPMENT FOR SPACECRAFT THE DEVELOPMENT OF ONBOARD MANEUVER PLANNING ALGORITHMS TENDS TO BE HINDERED BY A SIGNIFICANT CHALLENGE: THESE ALGORITHMS MUST BE ABLE TO EFFICIENTLY DESIGN MANEUVERS TO BE FEASIBLE SAFE AND ROBUST TO UNCERTAINTIES WHILE ACHIEVING A VARIETY OF SHORT-TERM AND LONG-TERM GOALS. IN THE ABSENCE OF WELL-DEFINED ANALYTICAL DESCRIPTIONS OF THESE REQUIREMENTS THIS CHALLENGE HAS PREVIOUSLY ONLY BEEN WELL-SOLVED BY HUMANS WHO POSSESS THE ABILITY TO REASON LEARN AND ADAPT. DURING SPACE-BASED OPERATIONS ORBITAL MANEUVERS ARE REGULARLY REQUIRED FOR STATION-KEEPING RECONFIGURATION COLLISION AVOIDANCE AND RENDEZVOUS. IN EACH FUNDAMENTAL OPERATIONAL MODE THE MANEUVERS ARE IMPLEMENTED OVER DIFFERENT TIME SCALES: WHILE STATIONKEEPING OPERATIONS ARE TYPICALLY PERFORMED BY GROUND-BASED TEAMS OVER LONG TIME INTERVALS UNEXPECTED HAZARDOUS OBJECTS MAY NECESSITATE COLLISION AVOIDANCE MANEUVERS OR COMPLETE ORBITAL RECONFIGURATION TO BE IMPLEMENTED IMMEDIATELY. FURTHERMORE FOR A FRACTIONATED SPACE-BASED SYSTEM AN INCREASE IN THE NUMBER OF SPACECRAFT OR COMPONENTS RESULTS IN THE STATION-KEEPING OPERATIONS PLANNED BY HUMANS BECOMING MORE COSTLY AND TIME-CONSUMING TO MANAGE. TO ADDRESS THESE CHALLENGES SPACECRAFT WILL NEED TO BE ABLE TO AUTONOMOUSLY MANEUVER WITHOUT RELIANCE ON GROUND-BASED HUMAN INSTRUCTIONS FOR EITHER ROUTINE OR IMMEDIATE ACTIONS. WE PROPOSE TO INCORPORATE THE CAPABILITIES OF HUMANS TO LEARN AND ADAPT INTO THE MANEUVER PLANNING PROCESS. THE PROPOSED WORK FOCUSES ON DEVELOPING A NEW TECHNOLOGY FOR AUTONOMOUS ONBOARD PLANNING OF SPACECRAFT MANEUVERS IN CHAOTIC DYNAMICAL REGIMES VIA MACHINE LEARNING. TO ACHIEVE THIS GOAL THE FOLLOWING OBJECTIVES WILL BE COMPLETED: 1) OFFLINE LEARNING FOR FEASIBLE AND EFFICIENT MANEUVERS: DEVELOP AND VALIDATE A MACHINE LEARNING ALGORITHM FOR MANEUVER PLANNING IN A CHAOTIC GRAVITATIONAL ENVIRONMENT ASSUMING FULL STATE KNOWLEDGE. INVERSE REINFORCEMENT LEARNING WILL BE USED TO UNCOVER THE REWARD FUNCTION BEING OPTIMIZED BY A HUMAN FLIGHT DYNAMICIST WHO IS ABLE TO BALANCE A VARIETY OF GOALS THAT CANNOT BE DESCRIBED ANALYTICALLY. A REINFORCEMENT LEARNING ALGORITHM WILL THEN LEARN TO DESIGN THE MANEUVERS THAT SAFELY AND EFFICIENTLY OPTIMIZE THIS REWARD FUNCTION DURING STATION-KEEPING AND ORBIT RECONFIGURATION. 2) ONLINE LEARNING FOR ROBUSTNESS AND RESILIENCY: ADAPT AND CHARACTERIZE THE CAPABILITY OF THE REINFORCEMENT LEARNING ALGORITHM TO AUTONOMOUSLY AND RESILIENTLY PLAN MANEUVERS WITH INCOMPLETE AND UNCERTAIN STATE INFORMATION. IN A CHAOTIC SYSTEM WHERE SMALL UNCERTAINTIES OR INCOMPLETE INFORMATION CAN PROPAGATE TO PRODUCE LARGE ERRORS THE POLICIES AND MODELS LEARNED IN THE FIRST OBJECTIVE MUST BE UPDATED. 3) AUTONOMOUS INTROSPECTION AND MACHINE SELF-CONFIDENCE: DEVELOP TECHNIQUES TO ENABLE THE MACHINE LEARNING ALGORITHM TO HOLISTICALLY EXAMINE ITS OWN PERFORMANCE ADJUST APPROPRIATELY AND COMMUNICATE AND IMPROVE ITS SELF-CONFIDENCE TO ENSURE EFFICIENT SAFE AND TRANSPARENT OPERATION. BY CONTINUALLY MONITORING MANEUVERS AND THEIR SUCCESS IN ACHIEVING THE DESIRED GOALS THE USE OF A LEARNING ALGORITHM IS EXPECTED TO REDUCE THE COMPLEXITY AND RISK ASSOCIATED WITH IDENTIFYING LOWER-COST FEASIBLE AND ROBUST MANEUVERS ONBOARD A SPACECRAFT. THE DEVELOPED TECHNOLOGY CAN ENHANCE A MISSION BY REDUCING THE OPERATIONAL COST AND COMPLEXITY OR ENABLE A MISSION BY SUPPORTING RAPID RESPONSE TO UNCERTAIN ENVIRONMENTS WHEN GROUND-BASED APPROACHES TO MANEUVER PLANNING ARE INFEASIBLE.