Asymptotically Optimal Ergodic Coverage on Generalized Motion Fields

Code for the paper "Asymptotically Optimal Ergodic Coverage on Generalized Motion Fields". We present a trajectory planner for ergodic coverage over time-varying distributions defined by dynamic particle flows (e.g., cattle herds, ocean currents, vortex fields) using Maximum Mean Discrepancy (MMD).

---

Installation

conda create -n temp python=3.12
conda activate temp

# Core
pip install jax jaxlib jaxopt numpy matplotlib scipy

# Geometry (3D experiment)
pip install trimesh open3d

# Ocean / geo data
pip install cartopy cmocean cmcrameri imageio dill svgpath2mpl

---

Experiments

1. Cattle Feeding

A herd of cows flows toward feeding sinks under repulsion and damping dynamics. The ergodic agent plans a trajectory that covers the herd's evolving spatial distribution.

Generate the flow and save data:

cd cattle_feeding
conda run -n temp python cattle_flow.py

Run the ergodic planner:

conda run -n temp python make_traj.py

---

2. Vortex Flow

Particles orbit a vortex center with wall repulsion. The ergodic agent tracks the rotating distribution.

Generate the vortex flow:

cd vortex_flow
conda run -n temp python flow_animate.py

Run the ergodic planner:

conda run -n temp python make_traj.py

---

3. Whale Search (Gulf of Mexico)

Whale pods are advected through real HYCOM ocean current data. The agent plans an observation path that maximizes cumulative encounter probability over the mission horizon.

Run the ergodic planner:

cd whale_search
conda run -n temp python make_traj.py

Ocean current data (whale_search/flow_data/flow_data.pkl) must be present. The flow field is loaded from HYCOM Gulf of Mexico velocity snapshots.

---

License

MIT — see LICENSE.