Activity-Based Scheduling of Science Campaigns for the Rosetta Orbiter

Steve A. Chien, Gregg R. Rabideau, Daniel Tran, Martina Troesch, Joshua Doubleday, Federico Nespoli, Miguel Perez Ayucar, Marc Costa Sitja, Claire Vallat, Bernhard Geiger, Nico Altobelli, Manuel Fernandez, Fran Vallejo, Rafael Andres, Michael KueppersIJCAI 2015

Abstract

The Rosetta mission's escort phase at comet 67P/Churyumov-Gerasimenko required a sophisticated approach to science operations planning. This paper introduces ASPEN-RSSC (Activity-Based Scheduling of Science Campaigns), an automated system that adapts the ASPEN planning engine to the unique constraints of the Rosetta orbiter. The system successfully balances the requirements of 11 scientific instruments against rigid engineering and orbital limitations. Operational results demonstrate that ASPEN-RSSC can generate a comprehensive 32-day Medium Term Plan (MTP) in roughly 20 minutes, managing over 2,000 observations and 10,000 constraints.

Introduction

The Rosetta mission was one of the most ambitious deep-space endeavors ever attempted, requiring the spacecraft to orbit and escort a comet for over a year. This mission profile necessitated a continuous stream of observations from a diverse set of 11 instruments, each with its own pointing, power, and data requirements.

The scheduling challenge was immense. Instruments often had conflicting requirements: some needed to point at the nucleus, others at the coma, and others had to avoid bright objects like the sun to prevent sensor damage. Furthermore, resources like power were strictly limited, and data had to be stored in onboard "packet stores" before being downlinked to Earth during specific windows. Manual scheduling using spreadsheets proved insufficient due to the massive web of conflicting constraints and the need for rapid plan adjustments based on comet activity. ASPEN-RSSC was developed to address these bottlenecks by automating the generation of feasible mission plans.

Methodology

ASPEN-RSSC utilizes a priority-first constructive scheduling algorithm. Science goals are defined as "campaigns" via XML-driven rules, allowing for flexible updates to mission logic (e.g., repeating observations at specific intervals) without altering the core software. The system integrates with an Attitude Generator Module (AGM) to verify slew feasibility and a specialized Downlink Allocator (DALLOC) for data management.

A critical feature of the system is its ability to calculate and visualize "valid constraint intervals" across multiple dimensions, such as power, data, and attitude. By showing scientists exactly why a request failed—for instance, due to a thermal constraint at a specific time—the tool facilitates manual refinement of science goals.

Constraint visualization showing valid start times and opportunity windows.

The planning process follows a tiered flow, moving from abstract "skeletons" to automated ASPEN-RSSC plans, and finally to high-fidelity command sequences in the MAPPS system. This successive refinement ensures that high-level science intent is accurately translated into executable spacecraft commands.

Results and Operational Impact

The system was operationally deployed starting in January 2014 and used for MTPs 1 through 18. In a typical 32-day planning cycle, such as MTP6 (August 2014), ASPEN-RSSC successfully managed 58 science campaigns, 2,119 individual observations, and 2,130 spacecraft pointings.

A 32-day mission plan (MTP6) showing the complexity of scheduled campaigns.

The efficiency of the greedy heuristic scheduler allows a full 32-day plan to be generated in approximately 20 minutes. A key insight from the deployment was the utility of distinguishing between "Prime" observations and "Rider" observations. Prime observations dictate the spacecraft's orientation, while Rider observations "tag along" to collect data while the spacecraft is already pointing in a useful direction. This strategy significantly maximizes science return within the mission's tight power budget.

Conclusion

ASPEN-RSSC represents a significant advancement in automated mission planning. By transforming a labor-intensive manual process into a streamlined, error-resistant workflow, it enabled the Rosetta science team to manage over 10,000 individual constraints per plan. While the current greedy approach prioritizes speed over mathematical optimality, the system's ability to handle immense complexity and provide actionable feedback to scientists makes it a model for future long-duration deep-space missions.

Summary updated

BibTeX
@inproceedings{chien2015activity,
  author = {Steve A. Chien and Gregg R. Rabideau and Daniel Tran and Martina Troesch and Joshua Doubleday and Federico Nespoli and Miguel Perez Ayucar and Marc Costa Sitja and Claire Vallat and Bernhard Geiger and Nico Altobelli and Manuel Fernandez and Fran Vallejo and Rafael Andres and Michael Kueppers},
  title = {{Activity-Based Scheduling of Science Campaigns for the Rosetta Orbiter}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure