SolarSim

Physics-based simulation of wearable solar harvester signals for motion sensing.

Reference implementation for SolarSim: Physics-Based Simulation of Wearable Solar Harvester Signals for Motion Sensing.

Wearable solar harvesters can sense motion while they harvest energy, but every existing solar-sensing system needs real photovoltaic recordings collected per participant and per lighting environment. SolarSim removes that bottleneck: it synthesises PV traces directly from motion trajectories using a radiometric view-factor model, with no training and no paired recordings beyond a one-off calibration per environment.

It takes exactly two inputs:

1. A motion trajectory — timestamped 3-D cell positions and outward normals from any motion-capture system.
2. An illumination environment — the geometry, position, and orientation of every light source in the scene.

Because the radiometric kernel depends only on geometry, everything hardware-specific — cell efficiency and area, source intensity, spectral response — collapses into a single calibration scalar κ per emitter. Changing the room means editing a scene file. Changing the harvester means refitting κ. Neither requires retraining anything.

Single- and multi-emitter scenes are supported, with mixed geometries (circular, rectangular, spherical) composing additively under one calibration.

Install

pip install -e .

Python 3.9+. Depends only on NumPy, SciPy, pandas, and PyYAML.

Quick start

# Inspect a lighting configuration.
pysolarsim describe --scene configs/scene/baseline_circular_6000K_40klx.yaml

# Fit kappa once for an environment, from ~60 s of paired motion and PV.
pysolarsim calibrate --scene configs/scene/baseline_circular_6000K_40klx.yaml --mocap configs/mocap/leap_csv.yaml --input data/baseline_15participants/real --max-seconds 60 --out results/kappa.json

# Synthesise power for a whole dataset, mirroring the input tree.
pysolarsim simulate --config configs/run/baseline.yaml

# Score simulated traces against measured PV.
pysolarsim validate --scene configs/scene/geometry_rectangular_6000K_40klx.yaml --mocap configs/mocap/leap_csv.yaml --input data/geometry_rectangular_6000K_40klx --calibration-mode per_recording

--input accepts a single recording or a directory.

Python API

from pysolarsim import Scene, RecordingSpec, calibrate_scene, simulate_recording

scene = Scene.load("configs/scene/baseline_circular_6000K_40klx.yaml")
spec = RecordingSpec.load("configs/mocap/leap_csv.yaml")

fit = calibrate_scene(scene, ["data/baseline_15participants/real/P01/Free Move__Synchronized.csv"], spec, max_seconds=60)
result = simulate_recording("data/baseline_15participants/real/P02/Circle__Synchronized.csv", scene, spec, kappas=fit.kappas)

print(result.power_uW)   # simulated trace, microwatts
print(result.metrics)    # pearson, rmse, nrmse, ...

Describing your own setup

The illumination environment is a scene file. Add a second emitter and multi-emitter superposition happens automatically, with non-negative least squares fitting one κ per source.

units: mm
emitters:
  - id: overhead_disc
    shape: disc              # disc | rectangular | spherical
    position: [-159.81, 868.82, 168.23]
    normal: [0, -1, 0]       # outward; defaults to straight down
    dimensions: {diameter: 200.0}
    kappa: null              # null means "fit me"
integration:
  disc: {n_r: 20, n_phi: 20}
  rectangular: {n_u: 20, n_v: 20}
calibration:
  method: auto               # auto | scalar_ls | nnls
  mode: reference            # reference | per_recording | fixed

The motion input is a mocap config that maps your capture system's column names, units, and sign conventions onto the simulator's canonical trajectory. Supporting Vicon, OptiTrack, or a Xsens export is a config edit, not a code change.

columns:
  time: time
  position: [leap_palm_x, leap_palm_y, leap_palm_z]
  normal: [leap_palm_normal_x, leap_palm_normal_y, leap_palm_normal_z]
units: {position: mm, time: s}
normal_convention: negate    # Leap reports the palm normal; the cell faces the other way
sample_rate_hz: 100
power:                       # only needed for calibration and fidelity scoring
  mode: v_times_i
  columns: {voltage: v_dc, current: i_dc}
  units: w

configs/scene/ ships a scene for every illumination configuration evaluated in the paper: four source intensities, three colour temperatures, a rectangular emitter, and three multi-emitter layouts.

Repository layout

Path	Contents
pysolarsim/	The simulator package (details)
configs/scene/	One scene file per illumination configuration
configs/mocap/	Input adapters for motion-capture formats
configs/run/	Complete run configurations
tools/build_scenes.py	Regenerate scene files from recorded layout notes
tools/run_fidelity.py	Score fidelity across all configurations

Data

The recordings evaluated in the paper are included in data/: synchronised PV, IMU, and hand-trajectory measurements from 15 participants, collected under QUT Human Research Ethics approval #9665. Participants are identified only by pseudonymous codes (P01–P15).

Each configuration directory is named identically to its scene file, so data/<name>/ and configs/scene/<name>.yaml always correspond. See data/README.md for the full layout and column reference.

The code runs against any recording a mocap config can describe, not just these.

Tests

python -m pytest pysolarsim/tests -q

Citation

@inproceedings{ghalwash2026solarsim,
  title     = {SolarSim: Physics-Based Simulation of Wearable Solar Harvester
               Signals for Motion Sensing},
  author    = {Ghalwash, Yasien and Khamis, Abdelwahed and Khalifa, Sara and
               Sandhu, Moid and Jurdak, Raja},
  booktitle = {Proceedings of the ACM on Interactive, Mobile, Wearable and
               Ubiquitous Technologies},
  year      = {2026},
}

License

MIT — see LICENSE.