vrp: Driver break discussion

Hi. Firstly, apologies if this belongs on a mailing list (I looked but couldn’t find one) but I wanted to have a discussion about the handling of driver breaks.

I’ve used jsprit now for various projects for the last 6 years and I kicked off the discussion originally that got the first iteration of breaks running there. I’m running on the assumption that you have some familiarity with jsprit given the setup of this library, and it looks like you’re making great progress in some of the shortfalls experienced there.

The one thing that causes no end of problems is driver breaks and it appears that you’ve taken the same default as jsprit did - the break inherits the location of the previous, tangible job and your own docs

The problem with this is that breaks are non-trivial in the solution itself. Because of things like EU regulations it simply isn’t good enough in a solution to have them modelled as soft constraints because the solution breaks the law if the break jobs get kicked out. Across jsprit, OR-Tools (broken for ages), VROOM and this library, I cannot find an implementation that can adequately address this.

Fundamentally, this falls down on the location inheritance. If I need to travel 5 hours from the north of the UK down to Cornwall, on a 9-5 shift (for argument’s sake), with a break in the middle, it’s totally impossible for a solution to ever do this job unless there is some stepping-stone job on the way. Tough luck if you’re dispatching a full truck for a single order. The stark reality is that vehicles will park up to satisfy the Tachograph at an intermediate location, but this is not possible either with a forward- or a backward-looking location inheritance.

This has led me down all sorts of paths, with instantaneous services at the warehouse immediately before a break just to try fudge this location inheritance, amongst other hacks. They never properly work. I think we need to move away from breaks being considered as a job that can be inserted into a route and, instead, have something time-based.

I’m engaged with a number of large logistics clients with complex constraints, and I have a wrapper library that calls custom and open-source solvers interchangeably, so I can test a number of very different scenarios. Given that you’re so active in your development at the moment and the fact that you’re addressing issues with the solvers that came before, I’d love to assist in testing if you have any ideas on how this long-standing problem can be fixed. Have you had any thoughts in this space and is this something you’d like to explore?

Thanks,

Josh