TL;DR
This paper details the development and application of the LADEE Activity Scheduling System (LASS), an AI-driven system designed to manage the complex activity planning for NASA's LADEE lunar mission. LASS successfully addressed the challenge of dynamically updating orbital constraints and coordinating multiple science teams, significantly improving efficiency and reducing errors in a time-critical operational environment.
What Problem Does This Paper Address?
Planning activities for a space mission like LADEE (Lunar Atmospheric Dust Environment Explorer) is incredibly complex. The mission involved multiple science instruments with specific observation needs, all of which had to be coordinated with the spacecraft's orbital path around the Moon. A major challenge was that the spacecraft's orbit was constantly being updated based on new tracking data. This meant that the precise times when certain orbital events (like passing over the lunar terminator) occurred were always in flux. These orbital events were critical for scheduling science observations, as many instruments had constraints tied to them. The planning process needed to be efficient enough to generate command sequences every other day, while also ensuring that all science and engineering constraints were met, and that the spacecraft's pointing accuracy requirements were satisfied. The paper addresses the need for an automated system that could handle this dynamic complexity and coordinate the efforts of geographically dispersed science teams and mission operations.
Methods
The core of the solution involved applying AI principles of problem decomposition and planning abstraction levels. This broke down the overall planning task into smaller, more manageable subproblems. The paper highlights the development of a mixed-initiative system called the LADEE Activity Scheduling System (LASS).
Key methodological innovations include:
- Orbital Event Representation: A novel approach was to introduce "orbital events" as a new activity type within the planning system. These events, like "Sunrise Terminator," were treated as special activities.
- Temporal Constraint Conversion: Science activity constraints that were tied to orbital events were converted into temporal constraints between the science activity and the corresponding orbital event activity.
- "Snap-to-Orbit" Mode: This feature in LASS allowed users to graphically drag generic orbital events and associated science activities, automatically snapping them to the nearest real orbital event and adjusting their timing to satisfy temporal constraints. This was crucial for handling the dynamic orbit updates.
- Mixed-Initiative Planning: LASS provided a mixed-initiative facility, allowing both automated conflict resolution and user intervention, balancing efficiency with human oversight.
- Activity Plan Abstraction Levels: The planning process was structured into six types of activity plans (mission, engineering skeletal, orbit allocation, instrument, strategic, tactical), each with a different temporal scope and level of detail. This allowed different teams to work independently at their respective abstraction levels.
Datasets
Not discussed in the paper. The paper focuses on the operational planning process and the system developed, rather than on training AI models with specific datasets.
Key Insights and Ideas
- Abstraction is Key to Complexity Management: Breaking down the planning problem into hierarchical levels of abstraction (e.g., orbit allocation plans) allowed geographically dispersed instrument teams to plan their activities independently without needing to constantly coordinate with each other or worry about the exact orbital timing. This significantly reduced the burden on the central planning team.
- Dynamic Orbit Constraints are a Major Hurdle: The paper underscores the difficulty of planning when key temporal constraints (orbital events) are constantly changing. The "Snap-to-Orbit" feature and the explicit modeling of orbital events were critical innovations to overcome this.
- Mixed-Initiative Systems are Powerful for Operations: LASS's mixed-initiative nature, combining automated constraint checking and resolution with user control, proved essential for balancing the need for speed with the requirement for accuracy and safety in a real-time mission environment.
- Generalizability of the Approach: The authors suggest that the method for managing dynamic orbital events is generalizable to other domains with similar characteristics, such as factory scheduling with dynamic delivery times.
Key Results
- Increased Efficiency: The LASS system significantly improved the efficiency of the activity planning process, reducing the time required for tactical planning. The lead tactical planner estimated the process would have taken at least twice as long without the automated update mechanisms.
- Reduced Errors and Conflicts: By automating constraint checking and providing tools for conflict resolution, LASS dramatically reduced human errors and conflicts among instrument teams' requests.
- Improved Coordination: The system facilitated asynchronous and independent work between science and mission planning teams, with activity plans serving as the primary communication medium.
- Successful Mission Support: LASS was used extensively by multiple teams throughout the LADEE mission, enabling the mission operations team to meet critical tactical deadlines.
Limitations and Future Work
- Manual Plan Modification for Communication Activities: The paper notes that while orbital event timing updates were automated, modifications to communication activities (due to changing view periods) still required manual intervention.
- Future Work on Communication Activity Automation: The authors suggest automating the modification of communication activities, potentially by introducing new event types or extending the constraint reasoning system (Dynamic Europa) to handle resource violations more broadly.
- Broader Application of Resource Violation Resolution: The paper points to the ongoing research in extending Dynamic Europa to handle resource violations, which could have broader benefits for activity planning beyond just orbital events.
Summary updated