Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision
Planning at 96 Million Kilometers from Earth: Challenges and Lessons
Nicola Muscettola
Computational Sciences Division
NASA Ames Research Center
Moffett Field, CA 94035-1000
Wednesday, March 1, 2000
refreshments 4:05PM, talk begins 4:15PM
TCseq201, Lecture Hall B
http://robotics.stanford.edu/ba-colloquium/
Abstract
On May 17th, 1999, the Planner/Scheduler (PS) of the Remote Agent
autonomy architecture became the first automatic planner to operate in
inter-planetary space. Over a period of 4 days the Remote Agent
controlled the Deep Space One spacecraft, the first mission of NASA's
New Millennium program. During the experiment the Remote Agent
received high-level goals from Earth. On-board the spacecraft PS
produced detailed plans that achieved them. PS did do so by using a
constraint-based model of the spacecraft activities and devices. The
produced plan was executed on board the spacecraft without any
intervention from ground controllers. Also Remote Agent demonstrated
the ability of detecting device faults and reacting to them while
keeping control of the spacecraft. In some fault scenarios PS was
called on to generate new plans that still achieved the original goals
while taking into account the degraded spacecraft capabilities.
On-board planning is a crucial capability for realizing NASA's vision
of a "virtual presence" in the universe, an armada of spacecraft and
rovers exploring celestial bodies in the Solar System and
beyond. Autonomy technology will boost the capabilities of ground
controllers to the point in which a single team will be able to
operate multiple missions simultaneously. On-board autonomy will
handle routine operations while ground controllers will be required
only during unanticipated off-nominal situations. Also, a spacecraft
that can recognize scientific phenomena, formulate new goals and
adjust its plans autonomously will significantly expand the kinds of
space exploration missions beyond those that are possible today.
In this talk we will discuss the PS component of the Remote Agent, its
constraint-based technology and the lessons learned during the Remote
Agent, particularly with respect to how to effectively use a plan
during reactive execution and how to validate a planner's
functionality in an operational, mission-critical application.
About the Speaker
Nicola Muscettola is lead of the Autonomy and Robotics Area in the
Computational Sciences Division of the NASA Ames Research Center. He
received both his Diploma di Laurea (B.S./M.S) and his Dottorato di
Ricerca (Ph.D.) from the Politecnico di Milano, Milano, Italy. Before
Ames he was a member of the research staff and then system scientist
at the Robotics Institute, Carnegie Mellon University. Nicola's
research interests are in planning, scheduling, reactive execution,
temporal reasoning, constraint propagation and planning-based method
for concurrent system verification.
bac-coordinators@cs.stanford.edu
Back to the Colloquium Page
Last modified: Mon Feb 28 12:00:22 PST 2000