Pre-Collision-Avoidance for LEO Spacecraft Using PACC-QMIX with an Orbital-Corridor Reward

Low Earth orbit is becoming increasingly congested, and hazardous conjunctions are creating a growing collision risk. Conventional collision avoidance relies on emergency maneuvers during the final warning window, leaving little time for decision-making, limiting maneuver opportunities, and consuming more fuel. To address these problems, this paper proposes a pre-emptive collision-avoidance framework that uses the longer period before an avoidance warning to plan small maneuvers in advance. These maneuvers move the spacecraft into a safe orbital corridor with a lower overall collision threat. We formulate pre-collision-avoidance as a multistage sequential decision-making problem and propose pre-action-component cooperative QMIX (PACC-QMIX). The maneuver decision is divided among three agents that determine the action components and one agent that determines the maneuver time. The shaped reward combines a normalized environment reward, an orbital-corridor reward, a maneuver-count penalty, and a separate collision-probability penalty. The collision-avoidance window is set to 1 day, while the pre-collision-avoidance planning windows are set to 7 and 14 days. Across ten randomized scenarios, the 7-day window increases the mean reward by 60.7% relative to the 1-day window, demonstrating the effectiveness of pre-collision-avoidance maneuvers. Under the 1-day and 7-day windows, PACC-QMIX achieves mean rewards of −0.0840 and −0.0330, respectively. Compared with MAPPO, Double DQN, and cross-entropy optimization, PACC-QMIX has the highest mean reward among the four methods in both windows. Five configurations are considered in the sensitivity experiments on the maneuver window and collision-probability threshold: <14-day, 1 × 10−8>, <14-day, 1 × 10−6>, <14-day, 1 × 10−4>, <7-day, 1 × 10−4>, and <1-day, 1 × 10−4>. The rewards for the 14-day window range from −0.0325 to −0.0276, while the reward for the seven-day window is −0.0330. These results show that the <7-day, 1 × 10−4> configuration provides a favorable balance among collision-risk control, fuel consumption, and pre-collision-avoidance lead time. For ten selected 7-day pre-collision-avoidance scenarios, stored maneuvers obtained in a two-body-only orbital environment are tested in an orbital environment including J2 perturbation and atmospheric drag. The experimental results show that the collision probabilities of all ten cases remain far below the threshold of 1 × 10−5 when the corresponding strategies are applied. In summary, PACC-QMIX achieves effective pre-collision-avoidance. By planning maneuvers in advance, it reduces fuel consumption, extends the orbital lifetime of satellites, and offers a new approach to the long-term stable operation of large-scale constellations.