RT23 |Vision-guided Autonomous Lunar Lander Simulation- a Hardware in Loop (HiL) Simulation

The study presented in this paper is an approach based on HiL Simulation to simulate the Powered Descent (PD) phase of a Lunar Lander using a representative physical model, a specified descent trajectory with velocity and attitude constraints and actual lunar imagery obtained from publicly available lunar image databases and other restricted sources . A 3-phase lunar landing is assumed consisting of a coasting Descend Orbit (DO), a Powered Descent (PD) and finally a Terminal Descent (TD) phase. It is further assumed that the lander is equipped with multiple sensors like the Inertial Measurement Unit (IMU), a surface tracking camera and Pulsed Laser Altimeter (PLA). The attitude and kinematic control of the lander is assumed to be guided by signals derived through sensor fusion .While the lander craft and its on-board controllers are simulated on an OpalRT OP4500 platform, the image processing hardware is interfaced through MATLAB. The final solution envisages a compact micro-controller based operation with an image processing FPGA and a processor core on the same board which is under development.. To eliminate the risk associated with manned lunar missions, autonomous lunar landing has emerged as the state of art in deep space exploration. Examples are the recent Chinese and Indian lunar expeditions. Autonomous Vision based hazard detection and avoidance system is one of the key research areas behind any autonomous landing mission for safe and soft landing on a planetary body and a comprehensive survey of these technologies is available in contemporary literature. The foremost amongst this is the NASA’s Autonomous Precision Landing and Hazard Avoidance Technology (ALHAT) . More recent developments include NASA’s Co-operative Blending of Autonomous Landing Technologies (COBALT) or the ESA’s Navigation for Planetary Approach and Landing (NPAL) use Vision Guided Control and Navigation (VGCN) which use real-time feature extraction for navigation of the space craft to designated co-ordinates (kinematic control) and also for correction of positional and attitude sensor data. A fantastic re-construction of the Change-3 lander’s actual descent trajectory is also available. The PD phase for the study presented in this paper starts at d 15 kms and during this phase the space-craft is subjected to deceleration in the horizontal plane as well as in the vertical direction by firing its thrusters. . The controller is a Hybrid Finite-State-Machine (FSM) based adaptive Proportional-Derivative (PD) controller and the velocity errors are corrected by adjustment of vertical thrust and attitude angles with suitable gain scheduling. The space craft relies on on-board IMUs and Terrain Relative Navigation (TRN) [10] using real-time feature extraction and matching across successive image so as to correct IMU errors. This phase , for the present study, is assumed to continue up to about 1.2km from the surface when the TD phase starts and the control set-points for navigation are generated at a faster rate i.e at 1 sec. interval for tighter position and velocity control. When the lander is within ± 1Km of its desired landing point in the horizontal plane and the velocities in the horizontal plane are less than ±10 cm/s, attitude stabilization is initiated so that the optical plane of the ground tracking camera is aligned with the central plane of the lander and the lander’s laser altimeters are switched on. The lander then descends to an altitude of approximately 140m and starts hovering using the on-board reaction wheel to create yaw around the z axis to detect minor hazards. If hazards are detected , the lander performs a re-targeting maneuver to perform a translation to a hazard free zone and touches down with a velocity of 1.8m/sec having engines switched off at 1m above the lunar surface. The entire simulation has been carried out on Opal RT with representative lunar data and imagery on an OP4500 platform.