Autonomous Rover Operations for Planetary Surface Exploration using Machine Learning Algorithms

Areas of Expertise: Dynamics, Estimation, Machine Learning The research goal of this NASA EPSCoR proposal is to develop the learning algorithms for an autonomous rover with an application of prospecting ice in extraterrestrial terrain. Our secondary objectives are improving understanding of our unique geologic assets, our rover, and our students/workforce. UH Manoa’s Institute of Geophysics and Planetology retains the finest planetary scientists but lacks the technologists and facilities required to independently design, build, and operate planetary surface exploration rovers. Autonomous exploration of deep space celestial surfaces, specifically the moon for water ice, is NASA’s second strategic goal (2.2) and fulfills advancement in NASA roadmap’s technological area (TA04). Regarding state-of-the-art, current rover autonomy algorithms are rigid in model formulation and inflexible to adaptation. Current machine learning architectures are powerful and adaptive but not human interpretable and offer minimal safety guarantees. This research incorporates machine learning into autonomous rover operations for exploring an unknown terrain with a formalized science objective while addressing safety and interpretability. The objectives of this proposal, led by Dr. Frankie Zhu, is to develop three algorithms that answer the following questions with the following contributions: (i) Given an encoded scientific goal, like quantifying ice, how does a surface agent autonomously explore an unknown environment that offers goal convergence and minimizes distance traveled? The first contribution is a reinforcement learning algorithm that optimizes science information gain and roving distance to generate an exploration strategy whilst also offering convergence guarantees. The science objective is to characterize the distribution of ice on a planetary surface utilizing a payload provided by Dr. Shuai Li (ii) What system identification method accurately learns a surface vehicle’s terramechanics model across various terrain whilst offering safe, interpretable, physics-informed predictions? A second contribution is to reveal the best terramechanics model structure for a reinforcement learning vehicle; models range from equations of motion derived from first principles, to a sparse set of fitted library functions, to neural networks, to symbolic regression. (iii) How does a rover independently and precisely estimate its own global position? The final contribution is a precise global position estimator based upon celestial navigation onboard the planetary surface vehicle with sensor fusion and a sensitivity study on sensor characteristics. With these three contributions, a robot imbued with a science objective can autonomously explore a planetary surface without orbiter or mission operations support. Our partners, NASA JPL and Astrobotic, can benefit from the ultimate research contributions of this body of work by taking the methodology as a whole or partitioning the algorithms selectively for full or partial autonomous operations. This project enables our team to raise the technology of this autonomous rover system from TRL 2 to 4. In implementation, the autonomous learning algorithm development will span two different environments: (i) performance validation of the learning algorithms in a computer-simulated environment and (ii) field tests in planetary surface analogues, including harsh volcanic fields akin to Lunar landscapes. In the process of implementing this research, young faculty will lead research staff, train students, arrange new facilities, and organize partnerships with NASA JPL and Astrobotic.

This is a mature technology (TRL 7+) — the realistic path in is usually NASA's Technology Transfer Program: licensing an existing NASA patent, or a Space Act Agreement to use NASA facilities/expertise directly. NASA also runs a startup licensing program with no upfront fee for companies formed to commercialize a specific NASA technology.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.