Dynamic Allocation of Cloud Resources to Parallelize Mars 2020 Flight Software Simulations for Sampling Operations and an Interactive Visualization Interface for Reviewing Results

Abstract:

Mars 2020 (M2020) aims to acquire and cache samples from the surface of Mars. The Qualification Model Dirty Testing campaign (QMDT) is focused on testing multiple sampling scenarios that could occur during the mission. Enabling this campaign is the Downselect pipeline and visualization frontend – a tool that simulates test configurations, checks sampling requirements, and visualizes the processed data in an interactive application.

The Downselect pipeline is deployed on Amazon Web Services (AWS) GovCloud using CloudFormation (CF) templates that define and allocate all services and resources. M2020 flight software (FSW) modules, which perform inverse kinematics for the robotic arm (RA) and separately evaluate drill placement and stabilizer articulation on localized topology, are isolated within unique Docker images deployed to AWS’s Elastic Container Service (ECS). Multiple containers associated with these images are spun up on demand to perform the parallelized evaluation of 3D points submitted to the pipeline. Results are written to AWS’s Relation Database Service, later used within the visualization interface where engineers can visualize processed rocks and filter locations based on the FSW evaluations to determine optimal sampling locations. These locations are output to a file used to generate RA sequences to physically sample the selected locations.

This architecture, specifically the dynamic allocation of resources via ECS, has enabled QMDT to have access to computational capability comparable to a supercomputer at a fraction of the price. With CF, the team has been able to create production and development instances of the tool instantaneously, and make account migration seamless.

The ability to evaluate multiple configurations in parallel has allowed M2020 engineers to increase their situational awareness during testing and understand the limitations of their test environment and the RA. The inclusion of this tool in QMDT's target selection process has parallelized operations and planning to enable faster implementation of selected targets, reducing the time for target selection from days to minutes.