Lisp at Remote agent experiment


Remote Agent

This experiment takes an even bigger step toward spacecraft autonomy with onboard computer software designed to make a wider variety of decisions. It is capa ble of planning and executing many onboard activities with only general direction from the ground.

The software is an autonomous "remote agent" of ground controllers in the sense that they rely on the agent to achieve particular goals. Ground controllers do not tell the agent exactly what to do at each instant; rather, they assign it more generalized tasks. The software package includes a inplanner/schedulerl. that generates a set of time-based and event-based activities, known as tokens, that are delivered to an inexecutivelh that is also a part of the software system. The executive makes decisions by taking into account knowledge of the spacecraft state, constraints on spacecraft operations and the high-level goals provided by the ground. The executive expands the tokens to a sequence of commands that are issued directly to the appropriate subsystems on the spacecraft. The executive monitors responses to these commands, and reissues or modifies them if the response is not what was planned.

Remote Agent™s design is flexible enough to handle a variety of unexpected situations onboard. Because of its access to a much more complete description of the spacecraft state than would be available to ground controllers in a traditional operations concept, it can make better use of onboard resources.

Remote Agent software was not designed to control Deep Space 1 throughout the mis sion; software was transmitted to the spacecraft after launch to control the ion engine and selected other systems during a 48-hour test period starting on the morning of May 17, 1999. On the morning of May 18, however, the experiment team detected an anomaly that interrupted execution of the experiment. The first indication occurred when Remote Agent did not com- mand the spacecraft's ion propulsion system to shut down as expected. While some portions of the Remote Agent software continued to operate, the component that issues commands had sus pended operation.

The spacecraft was determined to be safe and healthy. The experiment software appeared active and was allowed to continue processing while the experiment team and Deep Space 1 flight team analyzed the problem. After retrieving diagnostic data from the spacecraft, a ground command was issued that afternoon that halted the experiment.

By the time it was halted, the experiment had already achieved about 70 percent of its test objectives. A small bug in the very complex software was identified as the probable cause of the suspension, with the significant assistance of Remote Agent™s own self-diagnostic software. A successful follow-up experiment on May 21 completed the remaining objectives for the Remote Agent test. Presented with three simulated failures on the spacecraft, the experimen tal software correctly handled each event. The simulations included: a failed electronics unit, which Remote Agent fixed by reactivating the unit; a failed sensor providing false information, which Remote Agent recognized as unreliable and therefore correctly ignored; and a thruster stuck in the "off" position, which Remote Agent detected and for which it compensated by switching to a mode that did not depend on that thruster.

Remote Agent was developed by NASA™s Ames Research Center, Moffett Field, CA; JPL; and Carnegie Mellon University, Pittsburgh, PA.

The Remote Agent (RAX) development is based on Harlequin's LispWorks® product.

Remote Agent got NASA's 1999 Software of the Year award

Home    Previous    Up    Next
site map
Last update : 26/12/2000