Using World Models to Enable Safety Assurances for Uncertain High-Dimensional Systems

Autonomous systems are now frequently deployed into the real-world, so it is crucial that we can formally guarantee their safety and performance. However, doing so is especially challenging for systems whose dynamics are uncertain or complex, because one critical limitation of current formal verification methods is that the system dynamics must be known. For example, current tools cannot guarantee the safety of the Curiosity rover navigating unknown and complex terrain conditions on Mars. The lack of an effective formal safety framework forces Curiosity to operate at slow speeds and with frequent human intervention, bottlenecking the productivity of the rover. Inspired by recent progress in data-learned world models, this proposal aims to overcome current challenges of formal verification methods by using a learning-based world modeling approach to automatically discover an approximation of system dynamics. The learned approximation can then be used in formal methods to generate confident safety assurances that translate to guarantees on the real-world system. This could be an important first step towards enabling powerful safety tools to be used on space technologies, among other important real-world applications, for the first time.

This is early/mid-stage (TRL 2) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

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.