Abstract: With the improvement of space-borne computing capacity and emergency speed requirement of remote sensing constellation, it has become a key problem to generate observation plan autonomously and quickly for remote sensing satellite constellation. In order to improve the synergy and efficiency of autonomous mission planning in constellation, an adaptive distributed collaborative planning method is proposed. Based on the multi-agent method, the on-board autonomous management and inter-satellite mission cooperation mechanism and a heuristic task planning algorithm based on time window fitness are designed. Based on the S698PM embedded principle prototype, the simulation experiment shows that the Gap between the centralized planning model and the observation income is about 0.25 %, which is obviously better than other planning models in the aspects of planning time, load balancing and adaptability to large-scale problems.