DRILLAWAY: aDaptive, ResIlient Learning-enabLed oceAn World AutonomY (DRILLAWAY)

The objective of this proposal is to develop methods and demonstrations for autonomous, resilient, and efficient landed spacecraft operations on ocean worlds, with a particular focus on scooping, drilling and excavation tasks. We propose a holistic, end-to-end approach that combines vision, perception, control, and high-level planning with formal guarantees on system safety and performance. In order to tackle the inherent complexity and uncertainty in robotic operations with deformable media, we will adopt a hierarchical scheme with a loop of perception, model learning, and control on the lower level, and formal methods for certifiably resilient planning on the higher level. We will use geometric data from stereo vision and feedback from force sensors–as available in current NASA testbeds–to create models of the spacecraft's working environment and predictions of the outcomes of its actions, e.g., sampling yields, excavation depth, and spacecraft stability. We will expedite current methods for safe online model learning by exploiting all information available about the spacecraft's environment prior to or during the mission. For instance, our previous work showed that building a map between visual features and corresponding force sensor feedback significantly reduces the time needed for learning relevant dynamics. Additionally, we will use transfer learning, both in "sim2real" and "real2real" varieties, to translate models from previous experiments into an initial model estimate for the current mission. This approach will add value to our use of virtual and real testbeds during the period of performance: experiments on testbeds will serve not only to validate our algorithms, but also generate critical data to expedite learning in future missions. The learned models and observed environmental features will feed into the high-level planner, along with the measures of the learner's uncertainty about their correctness, obtained through introspective machine learning. By using the language of temporal logic, the planner will combine spacecraft's complex, multi-step tasks with safety specifications and express them as a reachability problem in a Markov decision process (MDP), thus intrinsically allowing for the possibility of undesirable events. In order to synthesize the appropriate high-level plans, the planner will use the framework of model checking combined with theories of resilient planning and guaranteed reachability developed in our prior NASA ESI project on extraterrestrial surface autonomy. Resilient planning will ensure that the spacecraft continues operating inasmuch as possible after sustaining damage, degradation, or partial loss of control over components, while guaranteed reachability enables determination of tasks that a degraded spacecraft can safely complete. We will adopt the framework of consumption MDPs for optimal energy management and prioritization of operations to maximize mission life. After synthesis, the high-level plan will be translated by the low-level controller into motion primitives, and implemented based on the learned models, while using the data collected during task completion to refine the models and reduce their uncertainty. We will validate our algorithms by leveraging both existing academic and NASA testbeds. In the early stages of work, we will complement the virtual OceanWATERS testbed with existing physical testbeds at the University of Illinois: UR5 and Franka Emika robotic arms to simulate excavation tasks, as well as the JPL-designed Robosimian quadruped robot to simulate disturbances on unstable terrain. These testbeds are equipped with 6- or 7DOF arms, force/torque sensors and depth sensors, and will allow us to simulate faults and unexpected events as well as collect data needed to perform transfer learning. Combined with testing on the OceanWATERS testbed, these environments will enable quick onboarding to NASA's physical testbed in the later stages of the project.

Developing Instrument or spacecraft technology to improve measurements for future planetary science missions

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.