Files
TIR_PROJ/tools/run_webots.sh
T
Johnny Fernandes dd5ac669e5 Webots sim-to-real fixes, DAgger pipeline, 360° proto variant
Today's session worked across the full Webots delivery stack — found and
fixed a cluster of bugs blocking the BC/RL transfer, then explored
training-side mitigations for the residual perception gap.

Bug fixes:
- Makefile FP_RATE default 2.0 → 0.0: BC demos used fp_rate=0 but RL
  fine-tune defaulted to fp_rate=2, poisoning the BC obs distribution
  and stalling PPO at 0% success across 1.46M+ steps.
- controllers/{shepherd_dog,sheep}/runtime.ini: Webots was launching
  controllers under system python3 (no numpy) and they were crashing
  silently. Pinned to the conda tir env.
- herding/config.py HERDING_WEBOTS preset: pen_latch_depth 0.5 → 2.0,
  max_new_tracks_per_step 3 → 1, static_reject 0.8 → 1.2. Stops phantom
  FPs near the gate from latching as permanently-penned tracks.
- herding/perception/sheep_tracker.py: penned tracks now decay at
  forget_steps × 8 instead of living forever. Adds get_positions
  min_freshness filter for deploy-time use.

Training/eval matches deployment:
- training/bc/collect.py: --dagger-policy flag for DAgger rollouts
  (policy drives, teacher labels) + --use-webots-preset for matched
  140° tracker + DR config.
- controllers/shepherd_dog/shepherd_dog.py: scan-fallback (0, 0.6) when
  BC/RL sees empty sheep_positions — recovers from FOV gaps.

Tooling:
- tools/dagger_round.sh: one-shot DAgger round (collect + concat + bc).
- tools/webots_sweep_gt.sh: full sweep with HERDING_USE_GT=1 for the
  perception-gap diagnosis matrix.
- protos/ShepherdDog360.proto: 360° FOV variant for the FOV-ablation
  comparison. Canonical proto stays at 140° per project spec.

Artifacts: v1 BC/RL policies for all 4 (drive × world) combos trained
in clean gym (success: diff/field 90-100%, diff/round 58%, mec/field
60-100%, mec/round 50-100%). DAgger r1/r2 BCs for diff/field show
12%→38% progression on gym HERDING_WEBOTS proxy but did not close
to actual Webots LiDAR (0/5 throughout). Next: LSTM policy or
learned tracker per the project-state memory.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-16 17:21:02 +00:00

178 lines
6.1 KiB
Bash
Executable File

#!/bin/bash
# Launch Webots with N sheep enabled and the chosen controller mode.
# Generates a temporary world file in worlds/field_test.wbt with sheep
# beyond N commented out, sets the env vars the dog controller reads,
# then execs Webots on it.
#
# Usage:
# tools/run_webots.sh [N] [MODE] [DRIVE] [WORLD]
# N : number of active sheep (1..10), default 10
# MODE : "bc" | "rl" | "strombom" | "sequential", default "bc"
# DRIVE : "differential" | "mecanum", default "differential"
# WORLD : base world name (without .wbt), default "field"
# Supported: "field" (rectangular), "field_round" (circular)
#
# Examples:
# tools/run_webots.sh 10 bc # behaviour-cloned MLP, diff drive
# tools/run_webots.sh 10 rl mecanum # KL-PPO fine-tune, mecanum wheels
# tools/run_webots.sh 5 sequential field_round # analytic baseline, round field
# tools/run_webots.sh 3 strombom mecanum field_round # Strömbom, mecanum, round
#
# Notes:
# * bc loads training/runs/bc/policy.zip, rl loads training/runs/rl.
# Override via HERDING_POLICY_DIR=/path/to/run env var.
# * Conda env "tir" must be active (provides stable-baselines3 + torch).
#
# Headless-ish (no 3D view, fast sim, no modal dialogs):
# WEBOTS_HEADLESS=1 make webots N=10 MODE=rl DRIVE=mecanum
# WEBOTS_HEADLESS=1 tools/run_webots.sh 10 rl mecanum
# This passes --no-rendering --minimize --mode=fast --batch to webots.
# Webots still needs a display (Qt); on a machine without one use e.g.:
# xvfb-run -a env WEBOTS_HEADLESS=1 tools/run_webots.sh 10 rl mecanum
# Optional extra CLI tokens (space-separated):
# WEBOTS_EXTRA_ARGS="--stdout --stderr" WEBOTS_HEADLESS=1 tools/run_webots.sh 10 rl
set -e
N=${1:-10}
MODE=${2:-bc}
DRIVE=${3:-differential}
WORLD=${4:-field}
if (( N < 0 || N > 10 )); then
echo "N must be 0..10, got $N" >&2; exit 1
fi
case "$MODE" in
bc|rl|strombom|sequential|universal|calibrate) ;;
*) echo "MODE must be bc|rl|strombom|sequential|universal|calibrate, got '$MODE'" >&2; exit 1 ;;
esac
case "$DRIVE" in
differential|mecanum) ;;
*) echo "DRIVE must be differential|mecanum, got '$DRIVE'" >&2; exit 1 ;;
esac
ROOT="$( cd "$( dirname "${BASH_SOURCE[0]}" )/.." && pwd )"
SRC="$ROOT/worlds/${WORLD}.wbt"
if [[ ! -f "$SRC" ]]; then
echo "World file not found: $SRC" >&2; exit 1
fi
DST="$ROOT/worlds/${WORLD}_test.wbt"
if [[ -n "${HERDING_POLICY_DIR:-}" ]]; then
RESOLVED_POLICY_DIR="$HERDING_POLICY_DIR"
else
# Try drive-mode-specific path first, then legacy path.
if [[ "$MODE" == "rl" ]]; then
DRIVED="$ROOT/training/runs/rl_${DRIVE}"
LEGACY="$ROOT/training/runs/rl"
else
DRIVED="$ROOT/training/runs/bc_${DRIVE}"
LEGACY="$ROOT/training/runs/bc"
fi
if [[ -d "$DRIVED" ]]; then
RESOLVED_POLICY_DIR="$DRIVED"
else
RESOLVED_POLICY_DIR="$LEGACY"
fi
fi
cp "$SRC" "$DST"
# Swap robot proto based on drive mode.
# Base worlds reference ShepherdDog (diff-drive). For mecanum we swap in
# ShepherdDogMecanum and inject mecanum contact properties.
if [[ "$DRIVE" == "mecanum" ]]; then
sed -i 's|"../protos/ShepherdDog.proto"|"../protos/ShepherdDogMecanum.proto"|' "$DST"
sed -i 's|^ShepherdDog {|ShepherdDogMecanum {|' "$DST"
# Inject mecanum contact properties into the contactProperties array.
# Strategy: find the closing ' ]' that ends the contactProperties block
# (it sits at 2-space indent, immediately before the WorldInfo closing brace)
# and insert just before it.
python3 -c "
with open('$DST', 'r') as f:
txt = f.read()
mec = ''' ContactProperties {
material1 \"MecanumWheel\"
coulombFriction [
1.0
]
bounce 0
forceDependentSlip [
0.01
]
softCFM 0.0001
}
'''
# The contactProperties array closes with ' ]\n}' (2-space indent ] then WorldInfo }).
# Insert the new block just before that closing ].
txt = txt.replace('\n ]\n}', '\n' + mec + ' ]\n}', 1)
with open('$DST', 'w') as f:
f.write(txt)
"
fi
# Comment out sheep N+1..10 by prefixing the matching Sheep { ... } line.
for i in $(seq $((N+1)) 10); do
sed -i "s|^Sheep .* \"sheep${i}\".*|# &|" "$DST"
done
active=$(grep -c '^Sheep' "$DST")
echo "------------------------------------------------------------"
echo "World : $DST"
echo "Mode : $MODE"
echo "Drive : $DRIVE"
echo "Sheep : $active active"
echo "Policy dir : $RESOLVED_POLICY_DIR"
echo "------------------------------------------------------------"
# Webots strips HERDING_* env vars from controller subprocesses in some
# setups, so we also write a runtime config file the controller reads.
cat > "$ROOT/herding_runtime.cfg" <<EOF
HERDING_MODE=$MODE
HERDING_POLICY_DIR=$RESOLVED_POLICY_DIR
HERDING_DRIVE=$DRIVE
HERDING_WORLD=$WORLD
HERDING_USE_GT=${HERDING_USE_GT:-0}
EOF
export HERDING_MODE="$MODE"
export HERDING_POLICY_DIR="$RESOLVED_POLICY_DIR"
export HERDING_DRIVE="$DRIVE"
export HERDING_WORLD="$WORLD"
# The controller writes this sentinel when all GT sheep are penned. We
# poll for it and kill Webots so the run finishes cleanly instead of
# idling for minutes after the task is done.
DONE_FILE="$ROOT/training/.run_done"
mkdir -p "$(dirname "$DONE_FILE")"
rm -f "$DONE_FILE"
if [[ "${WEBOTS_HEADLESS:-}" == "1" ]]; then
echo "[run_webots] headless flags: --no-rendering --minimize --mode=fast --batch"
# shellcheck disable=SC2086
webots --no-rendering --minimize --mode=fast --batch ${WEBOTS_EXTRA_ARGS:-} "$DST" &
else
# shellcheck disable=SC2086
webots ${WEBOTS_EXTRA_ARGS:-} "$DST" &
fi
WEBOTS_PID=$!
cleanup() {
kill "$WEBOTS_PID" 2>/dev/null || true
wait "$WEBOTS_PID" 2>/dev/null || true
exit 0
}
trap cleanup INT TERM
# Poll for the sentinel; bail when Webots exits on its own or when the
# user closes the window.
while kill -0 "$WEBOTS_PID" 2>/dev/null; do
if [[ -f "$DONE_FILE" ]]; then
echo "[run_webots] all sheep penned — closing Webots"
sleep 1 # let the controller print its line
kill "$WEBOTS_PID" 2>/dev/null || true
break
fi
sleep 1
done
wait "$WEBOTS_PID" 2>/dev/null || true