#!/usr/bin/env bash # Usage: run_strict_task_collection.sh TASK GPU COUNT OUTPUT [SEED_START] [WORKERS] set -euo pipefail task="$1"; gpu="$2"; count="$3"; output="$4"; seed_start="${5:-300000}"; workers="${6:-1}" export CUDA_VISIBLE_DEVICES="$gpu" OMP_NUM_THREADS="${OMP_NUM_THREADS:-10}" MKL_NUM_THREADS="${MKL_NUM_THREADS:-10}" export ROBOFACTORY_WRIST_WIDTH=640 ROBOFACTORY_WRIST_HEIGHT=480 cd /workspace/RoboFactory # RoboFactory resolves scene paths relative to its Python package while the # official asset snapshot is downloaded to the repository root. Fail early if # this explicit bridge or the required table mesh is absent. if [[ ! -e robofactory/assets ]]; then ln -s ../assets robofactory/assets fi test -f robofactory/assets/scenes/table/table.glb exec /workspace/venvs/robofactory-act/bin/python -u /workspace/act_liftbarrier/collect_wrist_task_dataset.py \ --task "$task" --count "$count" --workers "$workers" --seed-start "$seed_start" --gpus 1 \ --camera-width 640 --camera-height 480 --obs-mode rgbd --output "$output"