Tests¶

日本語 · README

Without the simulator¶

These run in CI (.github/workflows/ci.yml) on every push and pull request:

uvx ruff@0.17.1 check                                   # lint
uv run --no-project --with pytest --with numpy --with pyyaml --with pillow python -m pytest
                                                        # tests/unit: config, drive, power, IMU, URDF, STL, maps
docker run --rm kachaka-isaac-sim-ros:jazzy bash -c \
    'colcon test --packages-select kachaka_sim_bringup && colcon test-result --verbose'
docker run --rm -v $PWD:/repo kachaka-isaac-sim-ros:jazzy python3 /repo/tests/nav2_config_check.py
python3 scripts/build_usd.py --urdf-only                # build/kachaka.urdf, for the next one
docker run --rm -v $PWD:/repo kachaka-isaac-sim-ros:jazzy python3 /repo/tests/api_fake_check.py
docker run --rm -v $PWD:/repo kachaka-isaac-sim-ros:jazzy python3 /repo/tests/speed_limit_check.py
uv run --no-project --with pyyaml python scripts/run_scenarios.py scenarios/fake_room.yaml --backend fake

The unit tests on the URDF need scripts/fetch_assets.py first. tests/nav2_config_check.py starts slam.launch.py (with Nav2) and then navigation.launch.py on the saved map, with fake_robot (a point robot in an empty room, ros2 run kachaka_sim_bringup fake_robot) in place of the simulator. It checks that every lifecycle manager comes up and that Nav2 goals succeed.

tests/speed_limit_check.py measures the top speeds of manual and autonomous commands with and without a docked shelf in scripts/fake_sim.py (add --external to check a running simulator started with --shelf 0 0 0 --no-people).

tests/api_fake_check.py runs tests/api_check.py --no-cameras (the kachaka-api regression check: move_to_pose, rotate_in_place, return_home, shelves, error codes, history and more) against scripts/fake_sim.py, with SLAM, Nav2 and the command server as for the simulator. fake_sim.py runs the simulator's own ROS node and gRPC server, and replaces Isaac Sim with a kinematic robot in a 2D copy of the default room: the LiDAR sees its walls, boxes, people and shelf legs, the charger and the shelves work, and the robot stops at whatever it would run into. Add -e FAKE_SIM_ARGS=--trace to log the robot's pose and command once a second (in fake_sim.log under CHECK_LOG_DIR). It has no physics and no cameras, so these still need the simulator: camera images and object detection, the cliff and collision detection, how the robot moves under physics (wheel slip, bumping into things, pushing a shelf), Isaac scenes, and switching to a saved map (tests/api_map_check.py, navigation.launch.py).

CI has no GPU, so the simulator itself is checked locally: scripts/smoke_test.sh --nav starts it headless and runs tests/ros_smoke_check.py, then with SLAM + Nav2 the navigation, docking, shelf, command and object detection checks one after another in the same simulator run (each starts with sim/reset). A second run with a wall and a door runs tests/ros_door_check.py (Nav2 through the door, round it while it is closed, and through it again once it opens). A third run with layouts/room_auto_door.yaml runs tests/ros_auto_door_check.py (Nav2 through an automatic door that is never commanded: it closes behind the robot and opens as the robot comes back to it). A fourth run with layouts/room_movers.yaml runs tests/ros_nav_check.py --no-collisions (the Nav2 goals among people and a cart on the move, without touching them). It ends with tests/ros_hazard_check.py in a fifth run with a drop, and logs the collisions seen during the Nav2 checks to build/smoke_collisions.log.

With the simulator¶

With the simulator running, from the ROS 2 container (docker compose -f docker/compose.yaml run --rm -v $PWD/tests:/tests shell):

python3 /tests/ros_teleop_check.py   # drive the robot and print odometry
python3 /tests/ros_sensor_check.py   # topic rates, LiDAR ranges against the room walls, IMU
python3 /tests/ros_camera_check.py /tmp/images
python3 /tests/ros_nav_check.py      # Nav2 goals (needs slam.launch.py nav:=true)
python3 /tests/ros_nav_check.py --world office   # route in an Isaac scene (sim started with the same --world)
python3 /tests/ros_dock_check.py     # charger, battery and RETURN_HOME (same)
python3 /tests/ros_shelf_check.py    # shelf docking, towing and placing (same)
python3 /tests/ros_command_check.py  # MOVE_TO_POSE, MOVE_TO_LOCATION, SPEAK, goal_pose, cancel_all (same)
python3 /tests/ros_detection_check.py   # object_detection/result while turning in place (same)
python3 /tests/ros_hazard_check.py   # steps and collisions (no Nav2; sim started with --drop 1.0 0.0 0.4 0.8)
python3 /tests/ros_degradation_check.py   # degraded sensors (sim started with --config config/degraded.yaml --person 0.0 1.0 0.0)
python3 /tests/ros_door_check.py     # Nav2 and a door (sim started with --layout layouts/room_door.yaml)
python3 /tests/ros_auto_door_check.py   # Nav2 and an automatic door (sim started with --layout layouts/room_auto_door.yaml)
python3 /tests/ros_nav_check.py --no-collisions   # Nav2 goals among moving people (sim started with --layout layouts/room_movers.yaml)

The kachaka-api check runs on the host with the official client (same setup):

uv run --no-project --with kachaka-api==3.17.5.0 python tests/api_check.py
uv run --no-project --with kachaka-api==3.17.5.0 python tests/api_aio_check.py  # asyncio client and its streams

tests/api_map_check.py checks switch_map and set_robot_pose on a saved map; its docstring gives the steps (save the SLAM map, then restart the ROS 2 side with navigation.launch.py).

scripts/record_demo.sh records the GIFs in docs/images/ (scripts/record_demo.sh dock for the charger, shelf for the shelf, front_stop for the front obstacle stop, notebooks sensors|control|map for the demo notebooks in JupyterLab, with Chromium on the host); scripts/record_torch_demo.sh docs/images/torch.gif records the torch lights; scripts/record_dashboard_demo.py docs/images/dashboard.gif records the web dashboard (with the simulator, Nav2 and the dashboard running, and Chromium on the host), and with --scenario reset it records docs/images/sim_reset.gif. scripts/detection_demo.py --out docs/images/object_detection.gif records object detection (kachaka-api only).