A goal-directed agent for providing situation awareness to mission operators and crew members will facilitate safe, affordable and effective human and robotic exploration.
Automated Data Processing for Mission Operations
To safely and effectively carry out their tasks, mission operators and crew will need rapid access to relevant information. The traditional approach to this problem is too time consuming and is not affordable. The proposed technology provides a fast, adaptive and flexible mechanism for obtaining situation awareness.
We are developing agent technology for deriving data or information requested by users. In the context of an exploration mission, such agents will answer questions or requests for information posed by mission operators or astronauts in orbit, in a habitat or on EVA, e.g.:
Show me on a map the locations of these resources.
Which rover can get there the fastest?
What is the safest path to follow?
Show me the repair manual for this unit.
Such answers will enhance situation awareness and aid astronauts in carrying out tasks. Fulfilling these queries will require the agent to access and integrate data from many diverse sources, such as sensors, science instruments, databases, log files and activity plans. It may also require scheduling additional observations by rovers or other assets, and running software for simulation, 3D visualization or planning and scheduling.
Image left: Flow Chart Detailing Automated Data Processing.
When answering questions requires scheduling new data acquisition activities, the level of integration is raised even further. Tasks to acquire new data may compete with tasks that were previously scheduled, requiring replanning. A dialogue and negotiation capability will be needed to resolve conflicts and communicate costs
In contrast to the traditional approach of automating data processing by means of scripts, our agent-based approach provides the following advantages:
New data sources, data-processing operations or other capabilities can be added or removed, while the system is running, without the need to modify existing scripts.
Failures or unexpected events can be detected and recovered from to automatically. Anticipating all possible failures in a script or program is notoriously difficult.
The agent is not limited to a fixed set of goals. It is capable of using available resources to achieve requests that may not have been anticipated at design time.
This research capability builds on work developing IMAGEbot, an agent for processing Earth science data and ecological forecasting. IMAGEbot has been demonstrated in the Terrestrial Observation and Prediction System (TOPS), an ecological forecasting system that integrates many diverse data sources, including satellite data and weather station data. Requests to IMAGEbot are in the form of descriptions of desired data products. Given these requests, it invokes the appropriate remote operations on the TOPS server to generate the requested product and delivers them to the user.
IMAGEbot is based on a constraint-based planner specialized for data processing domains. A planner is a program that produces a plan
in response to a goal
, a set of conditions that the plan must bring about. The goals our planner accepts are descriptions of desired data products, and the plans it generates are dataflow programs
that generate the requested data. A dataflow program is composed of data-processing operations, each of which can have multiple inputs and outputs, in which outputs of one action can be connected to inputs of another.
NASA Ames is a world leader in automated planning and scheduling and constraint-based planning. The POC has over ten years of experience in the area of planner-based software agents.
Under current practices, substantial time and labor is required to obtain situation awareness from the large quantities of data and diverse software systems involved in a mission. Routine data processing often consumes over 80% of the labor required to analyze data, a substantial tax on both the cost and the time required to obtain information from the data. In emergency situations, the time required to obtain situation awareness could affect the safety of the mission and crew.
Image right: Complex Data Processing for Mission Operations.
To ensure safe, affordable and effective human or robotic exploration, it is important to automate as much of this data processing as possible. The current practice is to automate data processing using shell scripts or script-like programs, written before the start of the mission. Any required data products that were unanticipated before the mission began must be generated at a substantial cost in time and labor, resources that are usually in short supply. This difficulty in accessing and using data often results in dramatic under-utilization of data that were acquired at considerable cost.
Sustained human exploration of the moon and beyond under realistic budget constraints precludes relying on the huge standing armies of the past. This requires that we automate as much as possible. It is especially important to automate the conversion of data into high-quality information, in order to increase the effectiveness of the small operations teams of the future.