visualisation

Visualisation toolkit — compare N agents on one segment

Adapted from KB003/code/viz_compare_{rerun,plotly}.py. Given one segment of real driving data and any number of agents (each shipping a final-model/predict.py), overlays every agent's predicted yaw rate + integrated trajectory against the measured truth and the V0 kinematic baseline.

Setup

Needs Python 3.11+ and:

pip install numpy pandas plotly rerun-sdk

(rerun-sdk is only required if you want the 3D backend.)

Quick start

cd webinar-meta/visualisation

# 1. List every available segment (index, platform, device, route, idx)
python compare.py list
python compare.py list --platform HYUNDAI_IONIQ_5

# 2. Compare a handful of agents on segment #12 → plotly HTML
python compare.py 12 \
    --agent module-1/agent-01 \
    --agent module-4.v2.01/agent-01

# 3. Same comparison, open the rerun 3D viewer live
python compare.py 12 \
    --agent module-1/agent-01 \
    --agent module-4.v2.01/agent-01 \
    --spawn

# 4. Save an .rrd for later sharing/reopening
python compare.py 12 --agent module-1/agent-01 --rrd

Outputs land under webinar-meta/visualisation/out/<segment_slug>/.

Agent specs

An agent is any directory under webinar-AI/ containing final-model/predict.py (or predict.py at the top level) exporting predict(sim_df, platform) -> DataFrame with at least yaw_rate_pred_rads. Optional: x_m, y_m (otherwise the runner re-integrates from yaw + measured speed).

Spec forms:

module-1/agent-01                          # path; legend = "module-1/agent-01"
module-4.v2.01/agent-05:M4 agent-05        # path:label
/abs/path/to/some-agent-root               # absolute path

The V0 baseline (the x_m/y_m/yaw_rate_pred_rads already in the CSV) and measured truth are added automatically; you do not need to list them.

Presets

A preset is a JSON file under presets/ capturing a reusable agent list (and optionally a default segment):

{
  "segment": "5",
  "agents": [
    "module-1/agent-01:M1 V1",
    "module-4.v2.01/agent-01:M4 flagship"
  ]
}

Run it:

python compare.py --preset cohort-flagships --html      # use preset's segment
python compare.py 7 --preset cohort-flagships --rrd     # override segment

Save the current invocation as a preset:

python compare.py 12 \
    --agent module-1/agent-01 \
    --agent module-4.v2.01/agent-01 \
    --save-preset my-comparison

Three presets ship by default:

• example.json — one agent from M1 + one from M4, segment 0.
• cohort-flagships.json — agent-01 from every module (M1, M2, M3, M4).
• m4-all-agents.json — all ten M4 agents on the same segment (set on CLI).

Files

File	Role
compare.py	Top-level CLI dispatcher (list, segment + agents).
_runner.py	Segment discovery, agent loader, trajectory integrator.
compare_plotly.py	Plotly interactive HTML backend (6-panel overlay).
compare_rerun.py	rerun.io 3D + telemetry backend.
presets/	Saved comparison configs (JSON).
out/<seg_slug>/	Rendered HTML / .rrd / PNG.