Autonomous AI-Enhanced UAV System for High-Resolution Gravimetric Surveying Along Mars' Equatorial Zone: Design and Mission Implementation
- Paper ID
97728
- author
- company
Tsinghua University
- country
China
- year
2025
- abstract
This study presents a novel mission architecture combining autonomous aerial robotics and artificial intelligence (AI) to perform high-resolution gravity acceleration measurements along Mars' equator. Building upon advancements in Martian atmospheric modeling and rotorcraft flight dynamics, we developed a coaxial-rotor UAV capable of sustained flight in Mars' low-density atmosphere. A 1:3 scale prototype was tested in a Mars simulation chamber replicating atmospheric density, turbulence spectra, and diurnal temperature varying from $-73^\circ$C to $20^\circ$C, achieving stable hover at 12 m altitude with 15\% power margin. The AI navigation system employs deep reinforcement learning trained on 12,000 procedurally generated terrain models incorporating HiRISE-derived topography with 1 m/pixel resolution and MRO-derived gravitational anomalies. Field tests in Gebi Desert demonstrated 92\% success rate in autonomous obstacle avoidance across 15 km of flights over volcanic tuff formations. The proposed mission would deploy three such UAVs to conduct coordinated gravity surveys along $\pm1^\circ$ latitude of Mars' equator. Each vehicle carries a miniaturized accelerometer array with the sensitivity of $10^{-6}$ m/s$^2$ and laser altimeter synchronized with Mars Orbiter Laser Altimeter reference frames. Gravity measurements are taken at 5 km intervals, with positional accuracy maintained through SLAM-based visual odometry and Doppler updates from Mars Reconnaissance Orbiter. Each measurement site deploys a numbered radio-reflective marker containing a MEMS seismometer to enable future network-based interior structure studies. This approach addresses key limitations in current gravitational field models derived from orbital data of GMM-3B's 115 km resolution and localized surface measurements from InSight lander. Our simulations using JPL's Mars2020 terrain data show the system could map 1,200 km of equatorial transect per Earth year, detecting lateral density variations as small as 0.3\% in the upper 50 km crust. Technical innovations include: Adaptive rotor control algorithms compensating for Mars' 1-10 Hz turbulent eddies (cf. Richardson number analysis from MarsWRF simulations); Onboard AI prioritization of measurement sites based on real-time gravitational gradient analysis; Marker deployment mechanism optimized for $-100^\circ$C CO$_2$ ice regolith interactions. The mission architecture demonstrates how aerial platforms can bridge the resolution gap between orbital and surface-based planetary geophysical surveys.