AI-Driven Risk-Aware Scheduling for Active Debris Removal Missions

Abstract: The proliferation of debris in Low Earth Orbit (LEO) represents a significant threat to space sustainability and spacecraft safety. Active Debris Removal (ADR) has emerged as a promising approach to address this issue, utilising Orbital Transfer Vehicles (OTVs) to facilitate debris deorbiting, thereby reducing future collision risks. However, ADR missions are substantially complex, necessitating accurate planning to make the missions economically viable and technically effective. Moreover, these servicing missions require a high level of autonomous capability to plan under evolving orbital conditions and changing mission requirements. In this paper, an autonomous decision-planning model based on Deep Reinforcement Learning (DRL) is developed to train an OTV to plan optimal debris removal sequencing. It is shown that using the proposed framework, the agent can find optimal mission plans and learn to update the planning autonomously to include risk handling of debris with high collision risk.

• The paper introduces a DRL-based approach that formulates Active Debris Removal scheduling as a cost-constrained traveling salesman problem.
• It employs a dynamic state representation and a Bayesian-optimized DQN with a risk-sensitive reward system to prioritize high-risk debris.
• Validations using realistic collision data demonstrate the agent's ability to converge toward optimal solutions, improving mission safety and fuel efficiency.

AI-Driven Risk-Aware Scheduling for Active Debris Removal Missions

The paper explores the application of Deep Reinforcement Learning (DRL) in the domain of space technology, focusing specifically on scheduling for Active Debris Removal (ADR) missions in Low Earth Orbit (LEO). The necessity for such research arises from the increasing congestion in LEO due to space debris, which poses significant risks to both current and future space operations. This study addresses these issues by aiming to optimize the planning of ADR missions undertaken by Orbital Transfer Vehicles (OTVs) utilizing DRL-based methodologies.

Background and Problem Formulation

ADR missions in LEO involve multiple constraints and complex decision-making processes. The primary task is to deorbit several debris with a high level of autonomy while efficiently using time and fuel resources. The paper conceptualizes this problem as a Cost-Constrained Traveling Salesman Problem (CCTSP). Unlike classical methods that often rely on static optimization, the dynamic nature of debris management in LEO necessitates adaptive strategies that can respond to real-time changes.

Methodology

The proposed solution involves the implementation of a DRL agent to navigate the complexities of ADR missions. Key elements of this methodology include:

• State Space Representation: The state incorporates various mission parameters such as the remaining debris count, available fuel, mission time, the current debris location, and collision risk levels of the debris.
• Action Space and Transition: Actions are defined as selecting a piece of debris to deorbit next. State transitions do not depend on fixed time steps but on removal actions, updating factors like mission duration and fuel consumption.
• Reward System: The reward structure incentivizes the agent to reduce collision risk by prioritizing high-risk debris for deorbiting.

A DQN (Deep Q-Network) architecture supports the learning process, optimized through Bayesian techniques. This design is adapted to cope with the continuous state space presented by the real-time orbital data.

Results and Validation

The effectiveness of the DRL agent is validated using a dataset derived from the Iridium-Cosmos collision, allowing for a realistic simulation environment. An exhaustive search method validates the RL algorithm by benchmarking it against theoretical optimal solutions in a constrained problem setting. The results demonstrate the RL agent's ability to converge to this optimal solution.

Further, case studies are presented in which the agent is challenged to adapt to scenarios both with and without collision risk information. The findings indicate that providing risk prioritization allows the agent to make more efficient decisions, improving mission outcomes as evidenced by reward metrics.

Discussion and Implications

This research contributes significantly to the discourse on space sustainability by illustrating the feasibility of DRL for enhancing autonomous decision-making in ADR missions. The approach holds potential for real-time adaptation and optimization of OTV workflows, promising improvements in both safety and efficiency in space debris management.

Future research can be directed towards scaling the RL framework to accommodate larger datasets and more complex orbital mechanics, potentially integrating hybrid propulsion systems. Additionally, effective collaboration between space agencies in sharing real-time orbital data could augment the performance of learning-based agents.

In conclusion, the methodological advancements outlined in this paper offer a promising pathway in the pursuit of more sustainable and secure operations in LEO, paving the way for enhanced autonomous control in space mission planning through artificial intelligence.