Solving GcsTrajectoryOptimization with Minimum Duration Zero Leads to Cryptic Bug

[cohnt]

What happened?

If the user sets the minimum duration in regions to zero, this occasionally causes strange bugs in the construction of a PiecewiseTrajectory, when success would be expected. Here's a simple reproduction:

from pydrake.all import Point, Hyperrectangle, GcsTrajectoryOptimization

gcs = GcsTrajectoryOptimization(1)
start = gcs.AddRegions([Point([0])], 0)
goal = gcs.AddRegions([Point([1])], 0)
mid = gcs.AddRegions([Hyperrectangle([0], [1])], 1, h_min=0)

gcs.AddEdges(start, mid)
gcs.AddEdges(mid, goal)

gcs.SolvePath(start, goal)

Running this yields the following output:

INFO:drake:Solved GCS shortest path using Mosek with convex_relaxation=true and preprocessing=true and rounding.
INFO:drake:Finished 1 rounding solutions with Mosek, discarding 99 duplicate paths.
Failure at common/trajectories/piecewise_trajectory.cc:19 in PiecewiseTrajectory(): condition 'breaks_[i] - breaks_[i - 1] >= kEpsilonTime' failed.

I'll try to diagnose this and fix it sometime in the next couple of days.

Version

No response

What operating system are you using?

No response

What installation option are you using?

No response

Relevant log output

No response

[cohnt]

Looks like the issue is GCS produces a trajectory whose start and end times are equal (since it can travel at "infinite velocity"). What is the desired behavior here?

• Should we error if the user creates a region of order>0 with h_min=0?
• Should we catch this error and return a more intuitive error message to the user?
• Should we retime the trajectory to have a small duration, so the error doesn't occur?