SpaceTry: An LLM-Assisted Simulation Testbed for Space Mission Autonomy Engineering

Ricardo Diniz Caldas, Keila Lima, Roberto Natella, Patrizio Pelliccione

Keywords: space autonomy, self-adaptive systems, space robotics, behavior trees, runtime monitoring, testbed, Space ROS

Abstract: Autonomy is essential for planetary exploration because space robots must operate under uncertainty, limited supervision, and strong resource constraints while continuing to achieve mission goals. Because lab and field campaigns are costly and hard to iterate, in-house (in-vitro) testing can help screen scenarios before embodied validation. Existing robotics testbeds do not jointly provide this workflow for the space domain. We present SpaceTry, an open-source simulation-based testbed for scenario-based evaluation of autonomous space exploration missions. SpaceTry combines LLM-assisted scenario authoring from natural-language mission descriptions, configurable scenario artifacts for mission logic and monitoring, and a Space ROS/Gazebo runtime substrate for executing planetary-autonomy experiments. The framework does not prescribe any specific adaptation mechanism; instead, it exposes monitored violations as runtime events and leaves the response to the autonomy solution under evaluation. We describe the architecture, introduce a reference Mars outpost scenario extending the Space ROS NASA's Curiosity Rover Demo, and position SpaceTry as a testbed for comparing self-adaptive autonomy approaches before later lab or field testing.

Submission Number: 33