We have a simple test generator in TK branch which has an objective of covering of all model elements.
Traces are generated based on the following heuristics:
1) get a list of available transitions
2) get the number of times the edges in transitions were already executed
3) choose the least covered transition, or choose random if they are covered the same amount of times
4) execute the chosen transition, update the coverage statistics
5) goto 1.
I propose that the same heuristics could be used instead of just random walk in simulator. The advantage is that the algorithm above is simple (does not take much additional memory) and garantees that loops are avoided after they are exercised a few times.
The code is almost ready in TK branch, just needs to be properly integrated into Uppaal library and GUI. The statistics part also tracks the coverage of locations and variable values -- we can also easily provide a display of what elements have been covered by any trace.