Greg Detre
Tuesday, May 6 2003
Nicola Muscettola
Principal Scientist for Autonomy,
NASA Ames
Planning,
Execution, Life, the Universe, and Everything
One of the
most exciting endeavors pursued by human kind is the search for life in the
Solar System and the Universe at large. NASA is leading this effort by
designing, deploying and operating robotic systems that will reach planets,
planet moons, asteroids and comets searching for water, organic building blocks
and signs of past or present microbial life. None of these missions will be
achievable without substantial advances in the design, implementation and
validation of autonomous control agents. These agents must be capable of
robustly controlling a robotic explorer in a hostile environment with very
limited or no communication with Earth. In the first part, this talk gives an
overview of planned NASA planetary missions, with specific emphasis on Mars
exploration, and describes some of the requirements that these missions levy on
autonomous control agents. In the second part, the talk describes how some of
the research in real-time autonomous agents conducted at NASA Ames is
addressing these requirements.
This
research explores two main thrusts:
1) exploiting artificial intelligence planning as the core reasoning
engine of an autonomous agent;
2) building plans with flexibility over the
temporal and the resource dimensions.
About
Nicola Muscettola:
Dr. Nicola
Muscettola is Principal Scientist for Autonomy at the Computational Sciences
Division of the NASA Ames Research Center. Dr. Muscettola received all his
degrees from the Politecnico di Milano, Milano, Italy. He worked in planning
and scheduling research at Carnegie Mellon University from 1987 to 1993 where he
designed the Heuristic Scheduling Testbed System (HSTS). HSTS demonstrated
flexible temporal planning and scheduling on short-term planning for the Hubble
Space Telescope. In 1993 Dr. Muscettola joined NASA Ames. He was the architect
and project lead for the Planner/Scheduler module of the Deep Space 1 Remote
Agent that flew in May 1999. He is the architect of the Intelligent Distributed
Execution Agent, a re-engineering and rationalization of the Remote Agent
architecture, extending it to multi-agent system with real-time
guarantees.� Dr. Muscettola's research
interests are in automated planning and scheduling, temporal and resource
constraint propagation, multi-agent architectures, real-time control, and
validation and testing of autonomous systems.
The Mars lander so far have used depressingly little autonomy. For the most part, there's this horrendous, inflexible planning structure at Nasa that involves the scientists getting together every day (which is a Mars day, i.e. Slightlylonger than an Earth day, so everyone gets cranky and out of sync with daytime over the course a mission that lasts weeks) and decide upon the kinds of results they want, which get passed on to the engineers and manual rover operators, who may simply ignore the decrees if they're considered infeasible. Moreover, the thing is pretty much entirely open loop - that is, it tells mission control its current state, they plan an action, it performs it, and then waits for further instructions, with only minimal autonomous planning or navigation. The next one, planned originally for 2007 but postponed for at least a further couple of years is intended to be an order of magnitude larger. The one after that's going to be able to roam up to a few tens of kilometres, using a nuclear power source, and bring back a sample which will be put in a sealed cannister and will lift off to meet with an orbiter to be returned home.
He talked briefly about behaviour-basd robotics and subsumption architectures, and I noticed that Brian Williams seems to be at MIT and at Ames, but I couldn't tell exactly what lessons were being taken from there, or whether any of the legged robots have actually been sent on missions - it looks as though they've all been wheeled so far.