Vision-Based Dynamic Tasking for Earth-Observing Satellite Constellations

Unpredictable phenomena such as cloud coverage pose a challenge for conventional Earth observation. Long lead times can cause conditions to change between forecasting, scheduling, and image capture, leading to suboptimal observations. To combat this, dynamic tasking has been proposed as a mission concept, where perception and autonomy are moved onboard the spacecraft in order to quickly respond to changes in the spacecraft’s imaging schedule, leveraging the compute power available in modern Earth-observing spacecraft. In this work, we explore the use of a wide field-of-view body-fixed lookahead sensor, separate from the main imaging payload, to evaluate the real-time utility of upcoming imaging activities, specifically looking at the planning problem of scheduling imaging tasks combined with incorporating slews ahead of the orbit track to re-assess utility and re-optimize the schedule. We apply dynamic tasking with a lookahead instrument to the case of cloud avoidance for an Earth-observing satellite, extending to a leader-follower constellation of identical satellites. In this work we use areas of Europe as a simulation case due to the extremely dense cluster of imaging accesses requiring carefully balancing tradeoffs of performing the lookahead and optimizing the schedule against the opportunity costs of maneuvering the spacecraft. In our single-satellite case, throughput of cloud-free imagery improves by approximately 24%, and an improvement is found in 76 out of 100 simulated trials, even in perhaps the most challenging geographic region for this mission type.