End-to-end (E2E) pipeline integration test¶

Run bag record → convert/validate → ACT inference → ROS 2 / Rerun mock wiring in sequence. For equipment prep before connecting the real robot, see hardware.md.

Prerequisites¶

Docker runtime (ROS 2 execution environment)
Host Python virtualenv with pip install -e "port_pick_align/[training,dev]" installed
Environment split: the Docker image does not include PyTorch, so checkpoint inference runs on the host; inside Docker, run the visualization loop with policy_mode:=mock.

One-shot pipeline run¶

# If checkpoints/synthetic_act/best_model.pt already exists, skip retraining and smoke-test inference
./scripts/e2e_pipeline_smoke.sh

Stages:

Docker: record sensor topics from mock_sensor_publisher with ros2 bag record into data/e2e/recorded_bag.
Host: run scripts/convert_bag.py, convert to data/e2e/lerobot after validator quality checks (if the recorded bag has frame dropouts under load, continue with the fixed fixture sample_bag).
Host: inference smoke test with a trained checkpoint (--infer-only).
Docker: launch mock_inference.launch.py and verify topic I/O between the mock inference node and the Rerun node.

Single bag convert, integration test, checkpoint infer¶

1. Convert a single bag file¶

python scripts/convert_bag.py \
  --bag port_pick_align/tests/fixtures/sample_bag \
  --output data/from_bag/lerobot

2. Host integration test (CI integration equivalent)¶

(cd port_pick_align && pytest -m integration tests/test_e2e_bag_to_infer.py -q)

3. Checkpoint inference in a host ROS 2 environment¶

ros2 run pai_final_pkg act_policy_node --ros-args \
  -p policy_mode:=checkpoint \
  -p checkpoint_path:=/workspace/checkpoints/synthetic_act/best_model.pt

Feetech extra and udev prep¶

pip install -e "port_pick_align/[training,feetech,dev]"
# Linux: edit serial numbers in deploy/udev/99-feetech-so101.rules for your hardware and install under /etc/udev/rules.d/

For detailed hardware setup, see hardware.md.