v0.8.59 Stability Improvements

This fixes a bug where we were leaking built-in namespaces (like math) into the nimblephysics package.

v0.8.58 Creating MotionDynamicsDataset in Python

This release creates a PyTorch Dataset class that handles dynamically loading in on-disk binary motion dynamics data as needed, called MotionDynamicsDataset.

v0.8.57.1 Stability Improvements on x86_64

Removing ARM-only headers in SubjectOnDisk, which were breaking the build.

v0.8.57 Adding first draft of SubjectOnDisk binary seek()-able file format

This gets ready for training pipelines that use large amounts of motion data, and need to be able to page it in at random from disk to create batches.

v0.8.56 Fine tuning DynamicsFitter::timeSyncAndInitializePipeline()

Adding a zeroLinearResiduals() step DynamicsFitter::timeSyncAndInitializePipeline() to detect frames with missing forces.

v0.8.55 Anthropometric Prior Improvements

This changes the Anthropometric Prior to define its keypoints relative to meshes rather than bodies, to be compatible with a wider array of weird skeleton configurations. Also fixing an indexing bug in DynamicsFitter indexing.

v0.8.54 Stability Improvements

Minor fixes for:

• MarkerFixer::generateDataErrorsReport
• OpenSimParser
• DynamicsFitter::multimassZeroLinearResidualsOnCOMTrajectory

v0.8.52.1

Fixing a minor build error in v0.8.52

v0.8.52 Adding DynamicsFitter::multimassZeroLinearResidualsOnCOMTrajectory()

We can now fit multiple link masses simultaneously in a linear system, a generalization of DynamicsFitter::zeroLinearResidualsOnCOMTrajectory() that will tune individual link masses to get a trajectory closer to the desired.

v0.8.51 Stabilizing zeroLinearResidualsAndOptimizeAngular()

Eliminating GRF coverage blips in all calls that modify GRF coverage, stabilizing the smoothing on accs/grfs when detecting unmeasured external forces