REACT

Project Overview

The REACT project investigates the application of Deep Reinforcement Learning for satellite attitude control, addressing critical challenges in space systems autonomy. This research focuses on developing robust control policies that can handle both nominal operations and critical underactuated scenarios.

Conducted during my time at Argotec, this work demonstrates how RL-based controllers can be a fit solution for real systems, particularly in handling actuator failures and maintaining satellite pointing accuracy under challenging conditions.

Key Contributions

🚀 Novel RL Architecture

Used an adaptation of the PPO (Proximal Policy Optimization) algorithm, tailored for the unique challenges of satellite attitude control.

⚡ Robust Failure Handling

Created adaptive policies capable of automatically detecting and compensating for actuator failures without explicit failure mode programming.

📊 Comprehensive Evaluation

Extensive evaluations were run using both simulation and hardware-in-the-loop (HIL) testing environments.

🔧 Real-time Implementation

Demonstrated feasibility of RL-based control for real-time satellite operations in (HIL) Harware-In-the-Loop setup within space-graded on board computer.

Related Publications

Deep Reinforcement Learning Policies for Underactuated Satellite Attitude Control

Matteo El Hariry, Andrea Cini, Giacomo Mellone, Alessandro Balossino

NeurIPS 2021 Workshop on Deployable Decision Making in Embodied Systems

Reinforcement Learning for Underactuated Satellite Attitude Control