Reinforcement Learning for Planning and Task Coordination in a Swarm of CubeSats: Overcoming Processor Limitation Challenges

CubeSats have emerged as an innovative and cost-effective solution for space exploration and scientific research. Nonetheless, a significant drawback of CubeSats is their constrained computational capacity and memory. These processing limitations impose restrictions on the complexity of tasks that can be executed on-board, curtailing their capacity for intricate operations, such as data analysis and image processing. To address this challenge, one potential solution involves the utilization of a swarm of CubeSats, allowing them to pool their processing resources for executing intricate computations collaboratively. In this paper, we introduce an inventive approach for orchestrating planning and task execution within a CubeSat swarm, leveraging reinforcement learning techniques. Reinforcement learning, a subfield of machine learning, has demonstrated its effectiveness in solving diverse decision-making problems. In our proposed methodology, we employ a multi-agent reinforcement learning algorithm to harmonize the actions of multiple CubeSats. This algorithm is specifically designed to acquire optimal policies for each CubeSat based on the system's current state and the actions undertaken by fellow CubeSats. Our approach empowers the CubeSat swarm to share their processing capabilities, enabling the execution of complex computations that would be beyond the capabilities of a solitary CubeSat. To assess the efficacy of our approach, we conducted a series of experiments within a simulated environment. Additionally, we assessed the robustness of our algorithm in the face of environmental variations, demonstrating its continued effectiveness even in the presence of disruptive environmental factors. Our findings strongly suggest that employing reinforcement learning for planning and task coordination within a CubeSat swarm holds significant promise in surmounting the challenges posed by processing limitations. This novel approach has the potential to unlock new functionalities for CubeSats, thus broadening their applicability in scientific and commercial domains.