exotica
exotica copied to clipboard
ipab-slmc
Extensible Optimization Framework
Noticed while writing a unit test for a task map: - We load taskmaps (and many other things) as plugins where most of the relevant functions are provided via polymorphism...
The failure in the constrained optimisation stems from jumping and incorrect contact points, yet again:  We now see (a) offsets and (b) simply wrong locations.
Our new multi-OS CI does not allow modification of the source tree and thus cannot build documentation automatically. We should look into re-enabling building and deploying documentation from master.
The exposed write-able properties Q, H, W, Q_rate, H_rate, W_rate should be made private and exposed via getters/setters to a) check validity (dimensionality of matrices or value range) and b)...
We currently arent treating the antipodal equivalence of quaternions and minimise to the absolute target ``y`` rather than the minimum way to the orientation it actually represents.
As of August 2018, the following pairs of distance queries are not expected to work with ``CollisionSceneFCLLatest``: - [ ] Mesh vs Sphere - at distance [not implemented in FCL]...
With a fixed random seed, the computation time for obtaining the exactly same solution (because fixed random seed) increases _across_ restarts. A _reboot_ is required to resolve this.
After eliminating a few other factors (e.g. impact of "Range" on validity check between two states) it is clear that path simplification that we turn on by default eliminates required...
We should document whether a Ctrl+C will terminate on the next evaluation (i.e. may be even in a minor iteration or when evaluating gradients) or terminate to the next major...
I just noticed a cost difference for initial roll-outs between different solvers. The source of this is not taking into account the first timestep (for task cost), which we aren't...