The objective is to harness state-of-the-art explainable AI and operating system technology to build in an additional layer of dependability, accountability and intelligence between the critical core of a CubeSat and the environment it controls. Our initial evaluation found current operating systems deployed on CubeSats, e.g., FreeRTOS, are not fit for a future in which Launching States increasingly transfer liability for collisions resulting from failures or even cyber attacks to CubeSat Operators. We propose a novel solution that is verified, hence dependable, and which builds auditable real-time anomaly detection and overall health monitoring into the critical core of a CubeSat. More precisely, this project will result in a framework with the following contributions, which required us to balance the key requirements of verifiability and adaptability: (A) Trustworthy auditing of system health for post-disaster diagnostics. We will develop a method for automated root-cause analysis for disasters (e.g., tumbling and collisions), leveraging the diverse I/O used to maintain a Cubesat. We anticipate insurers will require such an analysis to resolve liability disputes, e.g., to prove that a fault is not due to negligence during CubeSat development and operations. (B) A hardware-isolated, dependable layer for autonomous disaster recovery. Instead of a set of hard-coded rules for disaster recovery, we will develop suitable AI mechanisms for onboard data analysis that take all I/O into account (not only telemetry). The framework must also adapt to unforeseen scenarios based on new data. Such a mechanism can make swift decisions in response to unexpected failures and perceived risks. Hardware-isolation ensures that monitors cannot be tampered with even when control software is compromised.
