THE OBJECTIVE OF THIS PROPOSAL IS TO DEVELOP THEORY ALGORITHMS AND DEMONSTRATIONS THAT (I) FORMALIZE SAFETY SPECIFICATIONS PLACED ON INTELLIGENT PHYSICAL SYSTEMS (IPS) IN SPACE AND GUARANTEE THAT AN INTELLIGENT SYSTEM REMAINS SAFE WHILE LEARNING ABOUT THE UNKNOWN ENVIRONMENT (II) INCREASE EFFICIENCY OF THE LEARNING CAPABILITIES OF IPS BY EXPLOITING PRIOR AND SIDE INFORMATION GAINED FROM PREVIOUS OR COMPLEMENTARY MISSIONS AND (III) ENSURE THAT IPS ARE RESILIENT TO SYSTEM DEGRADATION I.E. RETAIN THE ABILITY TO PERFORM THEIR MISSIONS INASMUCH AS POSSIBLE EVEN AFTER UNEXPECTED SIGNIFICANT DEGRADATION OF THEIR CAPABILITIES. THE PROPOSED RESEARCH WILL PROCEED IN THREE THRUSTS. THRUST I IS A FOUNDATIONAL APPROACH TO COMBINING THE NEED FOR IPS TO HAVE LEARNING CAPABILITIES NATURALLY REQUIRED WHEN OPERATING IN PARTLY UNKNOWN ENVIRONMENTS SUCH AS IN THE CASE OF A LUNAR OR PLANETARY ROVER WITH A CLEAR DEMAND FOR IPS TO COMPLETE THEIR TASKS AS SAFELY AS POSSIBLE. IN THIS THRUST WE WILL SEEK TO EXPRESS SAFETY SPECIFICATIONS THAT IPS NEED TO SATISFY WHILE COMPLETING ITS TASKS IN A FORMAL LANGUAGE OF TEMPORAL LOGIC AND BY BUILDING ON OUR PREVIOUS WORK ON REINFORCEMENT LEARNING WITH TEMPORAL LOGIC SPECIFICATIONS DEVELOP THEORETICAL AND ALGORITHMIC METHODS TO ENSURE SAFETY OF THE SYSTEM WHILE IT IS LEARNING ABOUT ITS ENVIRONMENT. THE EFFORT OF THRUST II WILL CONCENTRATE ON DEVELOPING CAPABILITIES THAT ALLOW IPS TO AUTONOMOUSLY REQUEST INTERPRET AND EXPLOIT ANY PRIOR OR SIDE INFORMATION AVAILABLE TO IT; NOTABLY INFORMATION OBTAINED FROM PREVIOUS OR COMPLEMENTARY MISSIONS. BUILDING UPON OUR PREVIOUS WORK ON EXPEDITING THE LEARNING OF TERRAIN DYNAMICS ON EXTRATERRESTRIAL SURFACES BY EXPLOITING ROUGH MAPS FROM ORBITERS WE WILL COMBINE FORMAL SPECIFICATIONS OF MISSION OBJECTIVES WITH PREVIOUSLY OBTAINED KNOWLEDGE TO PROMOTE REINFORCEMENT LEARNING STRATEGIES THAT MINIMIZE THE AMOUNT OF REDUNDANCY IN THE COLLECTED DATA AND COLLECT ONLY DATA IMPORTANT TO MISSION SUCCESS THUS EXPEDITING THE SYSTEM LEARNING AND ADAPTATION. THRUST III WILL SEEK TO ENSURE IPS RESILIENCE BY BUILDING UPON OUR CURRENT WORK THAT PROVIDES A FORMAL METHOD OF DESIGN OF AUTONOMOUS ROBOTIC SYSTEMS THAT ARE GUARANTEED TO REMAIN WITHIN A SAFE SET EVEN UNDER ACTUATOR MALFUNCTION. IN ADDITION TO BASIC SAFETY SPECIFICATIONS IT IS OF CENTRAL INTEREST TO ALLOW IPS TO RETAIN THEIR ABILITIES TO COMPLETE MORE GENERAL TASKS EVEN UNDER SENSING ERRORS OR PARTIAL LOSS OF ACTUATION. SIMPLY DEVELOPING STRATEGIES TO MITIGATE FAILURE ONCE IT OCCURS IS MYOPIC; IPS NEED TO BE A PRIORI DESIGNED IN A WAY THAT GUARANTEES THEIR ADAPTABILITY AND RESILIENCE UNDER A WIDE CLASS OF POTENTIAL ACTUATION OR SENSOR FAILURES. IN THRUST III WE INTEND TO COMBINE OUR WORK WITH METHODS OF FORMAL SPECIFICATIONS AND EFFICIENT LEARNING DEVELOPED IN PREVIOUS THRUSTS IN ORDER TO ENABLE IPS TO SUCCESSFULLY COMPLETE TASKS IN PARTLY UNKNOWN ENVIRONMENTS EVEN UNDER SIGNIFICANT ACTUATOR OR SENSOR MALFUNCTION. IN ADDITION TO THEORETICAL WORK ALL THREE THRUSTS WILL BE CONNECTED BY A COMMON CASE STUDY ON LUNAR ROVER OPERATIONS THUS DIRECTLY RESPONDING TO NASA'S NEEDS. THE CASE STUDY MODEL WILL INCLUDE FUNCTIONALITIES AND CONSTRAINTS TYPICAL FOR MISSIONS OF AUTONOMOUS ROVERS IN SPACE. WHILE FOCUSING ON LUNAR ROVERS FOR THE CASE STUDY WE EXPECT THAT THE MAJORITY OF DEVELOPED THEORY AND ALGORITHMS WILL BE APPLICABLE TO A WIDE CLASS OF ENVIRONMENTAL SETTINGS THUS SERVING TO A RANGE OF FUTURE MISSIONS.