978-1-4673-1813-6/13/$31.00 ©2013 IEEE 1 Probabilistic Hazard Detection for Autonomous Safe Landing Tonislav Ivanov Jet Propulsion Laboratory California Institute of Technology 4800 Oak Grove Dr. Pasadena, CA 91109 818-354-5017 tivanov@jpl.nasa.gov Andres Huertas Jet Propulsion Laboratory California Institute of Technology 4800 Oak Grove Dr. Pasadena, CA 91109 818-393-6695 huertas@jpl.nasa.gov Abstract — Future generation of landing craft will autonomously look at the surface during the terminal phase of powered descent and then, in real-time, choose and divert to a safe landing site in order to avoid hazards. Enabling technologies for such capability have been under development in recent years in the Autonomous Landing Hazard Avoidance Technology (ALHAT) project funded by NASA’s Exploration Technology Development Program. ALHAT is a comprehensive system that spans the approach and landing events – from de-orbit coasting to touchdown. In this paper, we focus on ALHAT’s perception task of detecting hazards in the sensed terrain and of selecting candidate safe sites for landing. This task, named Hazard Detection and Avoidance (HDA), occurs in the middle of the landing sequence. Our approach to HDA employs a probabilistic model in order to better manage the ubiquitous uncertainties associated with noisy sensor measurements and navigation. Also, we explicitly take into account the geometry of the lander and its interaction with the surface when assessing hazards. Experimental results on synthetic Lunar-like terrain show that our HDA algorithm can designate safe landing locations for a variety of terrain types and density and abundance of hazards. The complete ALHAT system is undergoing ground field-testing, and is scheduled for additional field tests on a one-hectare, lunar-like, hazard field recently constructed at NASA’s Kennedy Space Center (KSC). Although the focus of ALHAT is on autonomous planetary landings, a number of terrestrial applications can also benefit from out HDA system. TABLE OF CONTENTS 1. INTRODUCTION ................................................. 1 2. PREVIOUS WORK .............................................. 2 3. THE ALHAT SYSTEM ...................................... 3 4. HAZARD DETECTION ALGORITHM .................. 3 5. PROBABILISTIC MODEL ................................... 6 6. EXPERIMENTAL RESULTS ................................ 7 7. FUTURE WORK ................................................. 9 8. CONCLUSION................................................... 10 ACKNOWLEDGEMENTS....................................... 10 REFERENCES....................................................... 10 BIOGRAPHIES...................................................... 12 1. INTRODUCTION Landing of spacecraft requires autonomy. For unmanned missions, this need is essentially due to the time delay in communication between the spacecraft and operators on Earth. For manned missions, to the Moon for example, autonomy can assist astronauts when landing in dark or hazardous regions with minimal human supervision required. However, autonomous landing remains very challenging; it is probably the most critical engineering aspect of any mission. It is not for no reason that the final moments of Mars landings have been correctly dubbed “minutes of terror.” Thus, the development of safe landing capabilities, such as on-board HDA, is strongly desired. In fact, robotic landings with automated HDA are among the goals of NASA, other nations’ space agencies, and even private companies. Current efforts are underway in NASA’s ALHAT and Lander Vision System (LVS) projects to design, build, and validate HDA systems up to Technology Readiness Level (TRL) of 6 for future lunar and Martian missions respectively [11, 27]. A major hurdle to low-risk safe landing of autonomous craft has been a lack of reliable very high-resolution terrain observations, which are necessary to identify all hazards. For this reason, missions prior to 2008 have essentially landed “blind,” i.e. without full knowledge of lander-scale hazards from prior orbital observations and without the benefit of on-board hazard avoidance systems. More recently, advances in sub-meter imaging from high- resolution cameras on board reconnaissance orbiters have resulted in surface images with unprecedented detail. Notable are the HiRISE camera on board the Mars Reconnaissance Orbiter (MRO) with a nominal ground sampling distance (GSD) of 0.3m [28], and the Narrow Angle Cameras (NAC) on board the Lunar Reconnaissance Orbiter (LRO) with a nominal GSD of 0.5m [4]. Automated analysis of the images acquired by these electro-optical (EO) sensors have significantly improved landing risk assessment for recent Mars landed missions, namely Phoenix (PHX) [1, 10] and the Mars Science Laboratory (MSL) [11]. Similar analysis is possible for future lunar landed missions [16]. Nevertheless, even with superior orbital observations, PHX and MSL still required statistical approximations of unseen or undetected hazards from orbital analysis. Thus, added capability for on-board hazard detection will be advantageous and minimize risk even for missions whose landing sites have been mapped in high- resolution from orbit.