Scholarlink

<p>The proliferation of debris in low earth orbit poses significant risks to space safety and sustainability. Active debris removal (ADR), using orbital transfer vehicles (OTVs) to deorbit high-risk objects, offers a promising mitigation strategy. However, ADR missions require planning capabilities that balance economic viability with technical effectiveness, while adapting to evolving orbital conditions and shifting mission priorities. This paper develops a reinforcement learning framework based on deep Q-networks (DQN) that enables OTVs to plan debris removal sequences with high fuel efficiency while responding to changing risk priorities during mission execution. The framework models ADR planning as a Markov decision process with dynamic collision risk encoded directly in the state representation. Validation on realistic debris data demonstrates that the trained agent converges to optimal solutions, learns resource-efficient transfer strategies, and produces consistent mission trajectories. Agents with access to dynamic risk information achieve higher risk-weighted mission value compared to baseline agents, quantifying the operational benefit of adaptive planning. The results establish the feasibility of the proposed approach for risk-aware debris removal planning in future space traffic management operations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Risk-based mission planning for active debris removal using deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Pierre Nikitits,
  • Antoine Poupon,
  • Hugo de Rohan Willner,
  • Adam Abdin

摘要

The proliferation of debris in low earth orbit poses significant risks to space safety and sustainability. Active debris removal (ADR), using orbital transfer vehicles (OTVs) to deorbit high-risk objects, offers a promising mitigation strategy. However, ADR missions require planning capabilities that balance economic viability with technical effectiveness, while adapting to evolving orbital conditions and shifting mission priorities. This paper develops a reinforcement learning framework based on deep Q-networks (DQN) that enables OTVs to plan debris removal sequences with high fuel efficiency while responding to changing risk priorities during mission execution. The framework models ADR planning as a Markov decision process with dynamic collision risk encoded directly in the state representation. Validation on realistic debris data demonstrates that the trained agent converges to optimal solutions, learns resource-efficient transfer strategies, and produces consistent mission trajectories. Agents with access to dynamic risk information achieve higher risk-weighted mission value compared to baseline agents, quantifying the operational benefit of adaptive planning. The results establish the feasibility of the proposed approach for risk-aware debris removal planning in future space traffic management operations.