Efficient Task Network Generation

Background

Task Networks (TN) have been proposed as an alternative language for commanding spacecraft. However, specifying spacecraft tasks, their dependencies and ordering to achieve a goal is still largely a manual process and challenging to verify. As science goals become more complex, the large combinatorial space of possible tasks and sequences that the operator needs to consider is overwhelming and error-prone. Operators need a more efficient way to express goals and automated software to make TN generation more robust.

Problem

We are working to show that TN generation can be automated and made more efficient. The current approach is a manual process that uses a cumbersome input language. We are working to make writing TNs easier by making input more accessible to non-expert operators. We are also exploring the automated generation of task networks through the use of goal dictionaries.

Technology

We are using and exploring several technologies in our development process. One avenue of research was translating to and from the PLEXIL language. PLEXIL is a language for representing reactive plans for execution. We created a translator that can parse an MEXEC plan and translate it into an analogous PLEXIL plan.

For the automated generation of task networks, we have created a basic prototype using PDDL and a PDDL planner. In this formalism, simple goals can be used to create a task network by mapping the problem to PDDL.

Impact

This work enables more efficient operations by reducing plan creation time and increasing confidence on spacecraft commanding. It furthers JPL’s strategic goal for more capable autonomous robotic systems and mission operations.

Status

Project funded for Fiscal Year 2021 and 2022 by the NASA Center of Innovation Fund (CIF).

Publications

Team

Jet Propulsion Laboratory, California Institute of Technology
 Tiago Vaquero
 James Mason
 Vandi Verma

NASA Ames
 Jeremy Frank
 Michael Dalal

Sponsors

JPL Center Innovation Fund (CIF)