Planning and Scheduling
Discussion about Received Article
Richard Washington, Keith Golden, John Bresina:
Plan Execution, Monitoring, and Adaptation for Planetary Rovers
[message to moderator]
3.8.00 Anonymous Referee
Q1. Anonymous Referee 3. 8. 2000:
P2: "Given the varied state in which the rover can be, how
does the mode identification layer work? Details would be useful."
It was not clear how this piece worked and I would appreciate more details on how this is done. There are hints to the model based planning approach put forward by Brian Williams, is this the case? I would be interested in knowing how loops and iterations effect the models ability to realise it is back in the same state as before?
P2: "CX notices resources descrepancies, how does it
apportion blame when one or more activities are in conflict for the resource?"
It was not clear how the apportionment of blame was done, more details would be appreciated.
P2: "Compare and contract with execution architectures like Firby's
RAPS and Georgeffs PRS."
It is interesting that these systems were not compared. Is there a reason?
P4: "Sacrificing tasks is a context dependent thing and
this approach seems to be very fixed and rigid."
There seems to be a priority list of tasks to throw away or sacrifice given a fault, e.g. "low on power", "do not do science". Such a list is very context dependent and it was not clear how the sacrificing was carried out. More details would be appreciated. For example, the rover is running out of power and needs to get back into the sun light, having the motors work outside bounds might be the only way to have the rover survive?
P5: "How does the resource profiling work? More detail would be
appreciated, for example how does it compare with the Cesta and Smith
The profiling of resources is a difficult process and especially if the resource are ranges in both time and or level, e.g. "the experiment needs 5 to 8 watts of power between 0800 and 0815". More details would be appreciated.
P5: "The approach uses canned plans. What about a full replanning
capability, e.g. the 'Repairing plan on the fly' approach by
Drabble et al. in the first NASA Workshop for Planning and
Scheduling for Space?"
The paper hints that full repair would be too much of an overhead with actually stating in detail why. The approach described in the question has been applied to large scale logistics problems which have complex temporal and resource constraints. More of a comparison of the approaches would be appreciated.
P5: "How does the system deal with partial orders,
e.g. carrying out experimentation while the rover is moving,
Figure 2 seems to indicate this is not possible?"
I assume the approach works for linear sequences of actions? Could the approach be extended to partial orders or in fact two or more rovers working in a coordinated manner?
P6: "The approach seems to recover from failures once they have
occurred is their capability to handle future predicted
The examples indicate that the repair of problems takes place once they have occurred and there does not seem to be a predictive element e.g. if I continue to do this I will be in trouble?
P7: "What was the coverage of the contigent plans to situations the
rover faced e.g. 95% of the time the problem was repairable?"
It would have been nice if the paper had described the coverage of the contingent plans, i.e. what type of problem caused the rover to give up and ask for help.