Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LeLaR: The First In-Orbit Demonstration of an AI-Based Satellite Attitude Controller

Kirill Djebko, Tom Baumann, Erik Dilger, Frank Puppe, Sergio Montenegro

TL;DR

This work demonstrates the first in-orbit demonstration of an AI-based satellite attitude controller (LeLaR) trained entirely in simulation and deployed on the InnoCube 3U CubeSat. It details a split-policy DRL framework (RW/MT subnetworks) trained with SkipPPO, calibrated against a baseline PD controller, and protected by a Safety Cage to ensure safe operation. Despite Sim2Real gaps such as reaction-wheel dead time and residual dipole effects, LeLaR achieved steady-state pointing errors under 1 degree across multiple in-orbit maneuvers, validating autonomous, adaptive attitude control for small satellites. The results indicate strong potential for AI-driven autonomy in spacecraft, with future work aimed at momentum management, telemetry-driven simulation refinement, and broader applicability to other platforms.

Abstract

Attitude control is essential for many satellite missions. Classical controllers, however, are time-consuming to design and sensitive to model uncertainties and variations in operational boundary conditions. Deep Reinforcement Learning (DRL) offers a promising alternative by learning adaptive control strategies through autonomous interaction with a simulation environment. Overcoming the Sim2Real gap, which involves deploying an agent trained in simulation onto the real physical satellite, remains a significant challenge. In this work, we present the first successful in-orbit demonstration of an AI-based attitude controller for inertial pointing maneuvers. The controller was trained entirely in simulation and deployed to the InnoCube 3U nanosatellite, which was developed by the Julius-Maximilians-Universität Würzburg in cooperation with the Technische Universität Berlin, and launched in January 2025. We present the AI agent design, the methodology of the training procedure, the discrepancies between the simulation and the observed behavior of the real satellite, and a comparison of the AI-based attitude controller with the classical PD controller of InnoCube. Steady-state metrics confirm the robust performance of the AI-based controller during repeated in-orbit maneuvers.

LeLaR: The First In-Orbit Demonstration of an AI-Based Satellite Attitude Controller

TL;DR

This work demonstrates the first in-orbit demonstration of an AI-based satellite attitude controller (LeLaR) trained entirely in simulation and deployed on the InnoCube 3U CubeSat. It details a split-policy DRL framework (RW/MT subnetworks) trained with SkipPPO, calibrated against a baseline PD controller, and protected by a Safety Cage to ensure safe operation. Despite Sim2Real gaps such as reaction-wheel dead time and residual dipole effects, LeLaR achieved steady-state pointing errors under 1 degree across multiple in-orbit maneuvers, validating autonomous, adaptive attitude control for small satellites. The results indicate strong potential for AI-driven autonomy in spacecraft, with future work aimed at momentum management, telemetry-driven simulation refinement, and broader applicability to other platforms.

Abstract

Attitude control is essential for many satellite missions. Classical controllers, however, are time-consuming to design and sensitive to model uncertainties and variations in operational boundary conditions. Deep Reinforcement Learning (DRL) offers a promising alternative by learning adaptive control strategies through autonomous interaction with a simulation environment. Overcoming the Sim2Real gap, which involves deploying an agent trained in simulation onto the real physical satellite, remains a significant challenge. In this work, we present the first successful in-orbit demonstration of an AI-based attitude controller for inertial pointing maneuvers. The controller was trained entirely in simulation and deployed to the InnoCube 3U nanosatellite, which was developed by the Julius-Maximilians-Universität Würzburg in cooperation with the Technische Universität Berlin, and launched in January 2025. We present the AI agent design, the methodology of the training procedure, the discrepancies between the simulation and the observed behavior of the real satellite, and a comparison of the AI-based attitude controller with the classical PD controller of InnoCube. Steady-state metrics confirm the robust performance of the AI-based controller during repeated in-orbit maneuvers.
Paper Structure (31 sections, 30 equations, 27 figures, 29 tables)

This paper contains 31 sections, 30 equations, 27 figures, 29 tables.

Table of Contents

  1. Introduction and Background
  2. Methodology
  3. InnoCube Satellite
  4. InnoCube ADCS
  5. InnoCube ADCS software
  6. InnoCube PD controller
  7. Simulation Model
  8. Observation and Action Space
  9. Calibration
  10. Sensor Noise Model
  11. Moments of Inertia Estimation
  12. Residual Dipole Moment Estimation
  13. Base-Agent Training
  14. Base-Agent Reward Functions
  15. Subnetwork Policy
  16. ...and 16 more sections

Figures (27)

  • Figure 1: Overview of the components of the InnoCube satellite.
  • Figure 2: ADCS Software Modules.
  • Figure 3: InnoCube EQM in the Thermal Vacuum Chamber (open and exterior views).
  • Figure 4: First in-orbit maneuver of the base-agent. Maneuver duration from attitude control command (new att. command) to achieving $<$1$^\circ$ error on all axes (steady-state start) on 2025-10-30: 10:43:24 to 10:45:18 (114 s). The maximum wheel speeds were limited to [1000, 1000, 500] rpm and the maximum torque to [50, 50, 50] rpm/s via clipping.
  • Figure 5: Steady-state section of the first in-orbit maneuver of the base-agent from Figure \ref{['fig:maneuver_30.10.2025_10.40-10.50']}: Steady-state duration from 10:45:18 to 10:49:54 (276 s). The maximum wheel speeds were limited to [1000, 1000, 500] rpm and the maximum torque to [50, 50, 50] rpm/s via clipping.
  • ...and 22 more figures