Driver Breaks

[GuyBenhaim]

Hello,
For Google.OrTools.7.5.7466 in .Net, using Routing

I'm checking Breaks using IntervalVar. Aiming for one break per vehicle, that occurs anytime within a certain time-window.
With a small set (1 vehicle, 2 nodes: P + D) adding the breaks does not seem to change the solution times at all. I do get a solution.

I used the newer version of the API:

SetBreakIntervalsOfVehicle(
    std::vector<IntervalVar*> breaks,
    int vehicle,
    int pre_travel_evaluator,
    int post_travel_evaluator);

To simplify I set the last two parameters to -1 (which should make these return 0 for pre and post times).

Anyone successful with implementing such case? (i.e. getting a correct and consistent solution)
Can this API work, are we forced to use wacky fake-nodes?

p.s. Posted also in GitHub

Thanks

[Mizux]

did you try master ? v7.5 is quite old and IIRC there was some dev on break recently...

[GuyBenhaim]

Changed to v7.8 and used this:

break_intervals[vehicle] = model.solver().MakeFixedDurationIntervalVar(
                             0, end_time, 10000, false, "vehicle " + vehicle.ToString());
time_dimension.SetBreakIntervalsOfVehicle(break_intervals, vehicle, -1,-1);

With the SetBreakIntervalsOfVehicle condition solver seems to push the visits to be as late as possible, i.e. the solution always visits the end point at the latest possibe (end time = 60000). This occurs even when setting IntervalVar Duration=0.
However, commenting-out just this condition pulls the visits to be as early as possible and the end point is visited early.
The total time remains the same, the differences in visit times are the same ....

For example if visits with the condition are 57000 > 58000 > 59000 > 60000
Voiding it gives 0 > 1000 > 2000 > 3000

[GuyBenhaim]

Probing IntervalVar:
long startvalue = solution.StartValue(brk_var);
long endvalue = solution.EndValue(brk_var);

It is performed and gets start and end values assigned to it, but only pushes the visit times (i.e. no effect on time between node and total solution time )
Its gets more weird: setting span cost to 0, prevents the visits being pushed to the latest possible, but still no effect on visits and total time.
time_dimension.SetSpanCostCoefficientForVehicle(0, vehicle);

[GuyBenhaim]

I would like to know whether a basic one-break with one vehicle case works using time_dimension.SetBreakIntervalsOfVehicle
Any experience ? I did not yet mange to get this to work ... (v7.5 or v7.8)

[GuyBenhaim]

Is there a way for the vehicle break to start certain T time-units after the vehicle's start time, which is
IntVar start = time_dimension.CumulVar(routing.Start(v))

Either by setting the intervalvar's min_start and max_start parameters, using start:
break_intervals[vehicle] = model.solver().MakeFixedDurationIntervalVar(start, start+T, 10000, false, "vehicle " + vehicle.ToString());

Or by adding a proper constraint:
model.solver().MakeIntervalVarRelation(break_interval, start) ?
model.solver().Add(break_interval.StartMin() > start) ?
break_interval.SetStartMin(start) ?

[GuyBenhaim]

Any ideas? When vehicles start after t=0, defining breaks based on an absolute start time does not work ...

[GuyBenhaim]

Is this a feature to be supported? or a possible to trick to support on 7.X ?

[Mizux]

so using a duration dimension like explaining in another thread won't make it ?

[GuyBenhaim]

Should work as long as no is added Slack in the time-dimension.

BTW: I saw Laurent's earlier request. Should we move all discussion to GitHub?

[lperron]

move: If you want to. I will keep both systems in parallel to understand trends and preferences. Laurent Perron | Operations Research | lperron@google.com | (33) 1 42 68 53 00 Le ven. 4 déc. 2020 à 15:30, GMan <notifications@github.com> a écrit :

…

Should work as long as no is added Slack in the time-dimension. BTW: I saw Laurent's earlier request. Should we move all discussion to GitHub? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2193 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3ORBZYSEQSNPJXZX3LSTDW6PANCNFSM4SIL4VHQ> .

[Mizux]

Since break is a hot and specific topic, I like github issue since you can have several labels while discussion seems to be one category only. 😞
e.g.: https://github.com/google/or-tools/labels/routing_break it's nice to see all related routing break issues/topics
note: waiting for community/community#19 🤞