OpenCap to OpenSim to downstream analysis

library(PhysioMoCap)
#> Loading required package: PhysioCore

This vignette connects OpenCap (markerless motion capture from smartphone video) to OpenSim musculoskeletal computation and on to downstream kinematic/kinetic analysis. There are two routes, and they differ in whether they need a local OpenSim install.

Route A – use OpenCap’s cloud OpenSim output (self-contained)

OpenCap already runs OpenSim in the cloud: it scales a subject model and solves inverse kinematics, returning the joint angles as a .mot file. If you only need kinematics, pull that directly – no local OpenSim is required.

# Joint angles already computed by OpenCap's cloud OpenSim pipeline
pe <- readOpenCap("abcd1234-5678-90ab-cdef-1234567890ab", data_type = "kinematics")

# ... straight into downstream analysis
angles <- calculateJointAngles(pe)

Route B – run OpenSim locally (ID / muscle forces)

To go beyond kinematics – inverse dynamics (joint moments) or static optimization (muscle activations/forces) – you run OpenSim yourself. This is an optional path: it needs a working OpenSim backend and is off by default.

Enabling an OpenSim backend

PhysioOpenSim selects a backend automatically: a native library compiled in, or the opensim-cmd command-line tool. The native bridge is disabled by default so ordinary installs stay dependency-free; opt in one of these ways:

# (a) put the OpenSim CLI on your PATH (no rebuild) -- e.g. from a conda install:
#     mamba install -c opensim-org opensim
# (b) build PhysioOpenSim against an OpenSim SDK (configure auto-detects it):
#     OPENSIM_HOME=/opt/opensim-core R CMD INSTALL PhysioOpenSim

PhysioOpenSim::opensimDiagnostics()   # backend: "native", "cli", or "none"

From an OpenCap session, end to end

runOpenSimFromOpenCap() downloads the session’s scaled .osim model and marker .trc, then runs the requested tools. Inverse dynamics and static optimization need ground-reaction forces, which markerless OpenCap does not record – supply a measured or estimated OpenSim ExternalLoads XML via external_loads_file.

res <- runOpenSimFromOpenCap(
  session_id          = "abcd1234-5678-90ab-cdef-1234567890ab",
  tools               = c("ik", "id", "so"),
  external_loads_file = "walk_grf.xml"      # required for id/so
)

res$motion      # IK joint angles as a PhysioExperiment
res$outputs     # parsed .mot/.sto outputs (joint angles, moments, activations)

If you already have the model and markers locally (or want to reuse a model across trials), call the lower half directly:

model <- downloadOpenCapModel("abcd1234-5678-90ab-cdef-1234567890ab")

res <- runOpenSimFromMarkers(
  model_file          = model,
  trc_file            = "walk.trc",
  tools               = c("ik", "id", "so"),
  external_loads_file = "walk_grf.xml"
)

Pass dry_run = TRUE to write and inspect the OpenSim setup XMLs without a backend – useful for debugging or customising a template before running.

Downstream

The returned PhysioExperiment objects flow into the rest of the ecosystem:

angles  <- calculateJointAngles(res$motion)
moments <- inverseDynamics3D(res$motion)          # or read res$outputs$id
# ... and on to gait indices, or statistical parametric mapping in PhysioAnalysis:
# PhysioAnalysis::spmAnova(...)

Scope and honest limitations

• Ground reactions. OpenCap is video-only, so it has no force plates. Inverse dynamics / static optimization therefore require an external ExternalLoads file (measured GRF, or an estimate). Without one, only inverse kinematics runs; set require_external_loads = FALSE for kinematics-only inverse dynamics, which is valid only for non-contact phases (e.g. swing).
• Model scaling. PhysioOpenSim has no Scale tool; the subject-specific model comes from OpenCap (downloadOpenCapModel()).
• CMC / RRA. These need actuator and tracking-task files beyond what this wrapper infers. Run them with run_opensim_toolchain() together with PhysioOpenSim::opensimWriteCMCSetupFromTemplate() / opensimWriteRRASetupFromTemplate().
• CI. The actual OpenSim execution is exercised only where a backend is present; continuous integration installs no OpenSim, so those steps are skipped there by design.