Level 2 — Vrrrm! Now it's real test time¶

Same plan, real robots. The controller is now connected to a physical swarm through the Mari gateway over MQTT, and each DotBot runs the waypoint-following loop on board, localised by Lighthouse 2 (LH2). This is the ground-truth level: it's where real physics — turning radius, overshoot, localisation noise — meets the plan.

The driver, real_dotbot_pibt.py, is the conservative counterpart of the Level 1 script: it sends one waypoint per bot sequentially, then holds a synchronisation barrier that polls the bots' real LH2 positions until all have arrived. A bot that exceeds --step-timeout is logged and skipped instead of blocking the swarm (no deadlock).

Prerequisites¶

A provisioned DotBot swarm and a gateway flashed with the Mari firmware.
An LH2 setup (two base stations) calibrated for your arena — follow PyDotBot's Lighthouse 2 calibration guide.
The [calibrate] extra installed (see Installation).

1. Start the MQTT broker¶

mosquitto -c mosquitto.conf      # listens on :1883

2. Bring up the gateway and the bots¶

With the nRF gateway connected over USB, check the fleet:

dotbot swarm status              # should list the real DotBotV3 units

If the bots show as "Bootloader", they aren't running the DotBot application and won't appear in the controller. Flash and start the full sandbox app:

dotbot swarm flash <path-to>/dotbot-sandbox-dotbot-v3.bin -ys
dotbot swarm start
dotbot swarm status              # should now read "Running"

Note

Only the dotbot-sandbox-dotbot-v3 application provides LH2 advertisement and waypoint navigation. Demo apps like spin, move, or rgbled are not drivable from the controller.

3. Start the real controller¶

dotbot run controller \
    --conn mqtt://localhost:1883 \
    --swarm-id 1234 \
    --map-size 2000x2000

--map-size must cover your real LH2 area; --swarm-id is the one in the MQTT topic (/mari/1234/…). Do not run dotbot run simulator at the same time — it would occupy port 8000.

Confirm the controller sees the bots:

curl -s http://localhost:8000/controller/dotbots | python3 -m json.tool

You should see the real addresses with a non-null lh2_position and calibrated ≠ 0. An empty [] means the controller is still in simulator mode, or the bots are still in bootloader.

4. Run the PIBT test¶

Always dry-run first, then start small and scale up:

python real_dotbot_pibt.py --dry-run --seed 1     # no commands sent
python real_dotbot_pibt.py --seed 1 --steps 20    # real run

Option	Default	Role
`--threshold`	100 mm	Distance under which a bot counts as "arrived" at its cell
`--step-timeout`	8.0 s	Max wait per step; beyond it the bot is logged and skipped
`--settle`	0.3 s	Pause after arrival to let bots come to a stop
`--min-bots`	2	Localised bots required before starting
`--steps`	30	Maximum PIBT steps
`--seed`	random	Reproducible random goals
`--map-cells`	5	Grid resolution N×N (5 → 400 mm cells, 8 → 250 mm on a 2000×2000 map)
`--cell-mm`	derived	Cell size in mm (overrides `--map-cells`; centres at `gx*cell_mm + cell_mm//2`)

Reproducing the experiments¶

The batch harness runs many seeded trials at a fixed bot count and records per-run metrics and a summary under log/:

python real_dotbot_pibt_batch.py --bots 4 --runs 5
python real_dotbot_pibt_batch.py --bots 8           # 8 bots, 5 runs (default)

Each invocation writes:

a raw run log to log/raw_logs/,
per-run rows to log/l1_per_run.csv,
an aggregated log/l1_summary.csv.

The experiment sweeps cell size on a fixed 2000 × 2000 mm arena (4×4 / 5×5 / 8×8) and overlays the result against the Level 0 sweep.

Quick troubleshooting¶

Symptom	Likely cause
`curl` returns `[]`	Controller still in simulator mode, or bots in bootloader (`dotbot swarm start`).
`Connection refused`	Controller not running on `:8000`.
Bots not localised	LH2 coverage / calibration — recalibrate if positions look wrong.
One bot stalls a step	Expected — `--step-timeout` skips it after the delay and continues.