The Safe Trusted Autonomy for Responsible Space Program

TL;DR

The STARS study tackles safe and trusted autonomy for space by integrating reinforcement-learning-driven multi-satellite control, run-time assurance to enforce safety across 14 constraints, and human-autonomy teaming interfaces, all tested in a dedicated LINCS laboratory and validated on edge-space hardware. The approach combines neural-network control with ASIF-based run-time safety, and a directable HAI interface to support operator oversight and adaptability, enabling ground-trained policies to operate on lightweight spacecraft processors. Key contributions include a modular, open implementation of RTA (Safe Autonomy Run Time Assurance Framework), RL environments for docking and inspection tasks, and a validated terrestrial testbed that mimics orbital dynamics with real-time feedback. The work demonstrates that training on Earth and deploying at the edge is feasible within real-time constraints, offering a practical path toward safer autonomous space operations and a foundation for future operator studies to establish calibrated trust.

Abstract

The Safe Trusted Autonomy for Responsible Space (STARS) program aims to advance autonomy technologies for space by leveraging machine learning technologies while mitigating barriers to trust, such as uncertainty, opaqueness, brittleness, and inflexibility. This paper presents the achievements and lessons learned from the STARS program in integrating reinforcement learning-based multi-satellite control, run time assurance approaches, and flexible human-autonomy teaming interfaces, into a new integrated testing environment for collaborative autonomous satellite systems. The primary results describe analysis of the reinforcement learning multi-satellite control and run time assurance algorithms. These algorithms are integrated into a prototype human-autonomy interface using best practices from human-autonomy trust literature, however detailed analysis of the effectiveness is left to future work. References are provided with additional detailed results of individual experiments.

Figures (13)

• Figure 1: Human operator on-the-loop of a supervised autonomous satellite control concept.
• Figure 2: Reinforcement learning feedback loop.
• Figure 3: Relative motion Hill's reference frame in which the motion of deputy satellites are described with respect to a chief. The origin of the Hill's frame in centered on the chief spacecraft, $\hat{x}$ is aligned with the vector $\vec{r}$ pointing from the center of the Earth through the chief satellite, $\hat{y}$ is in the velocity direction of the chief satellite in its orbit, and $\hat{z}$ is aligned with the angular momentum vector of the chief in orbit about the Earth.
• Figure 4: Examples of torque-free rigid body dynamic behaviors. Shown are the body fixed axes as viewed in Hill's frame. Each example was generated over a different amount of time to exemplify distinctive qualitative features of each mode.
• Figure 5: Run time assurance control system architecture