diff --git a/docs/conf.py b/docs/conf.py
index 5e18c10..327abae 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -64,7 +64,7 @@
     ("py:class", "htss_rig_bluesky.devices.SampleStage"),
     ("py:class", "SampleStage"),
     ("py:class", "htss_rig_bluesky.plans.detector.Roi"),
-    ("py:class", "ophyd_async.epics.areadetector.aravis.AravisDetector"),
+    ("py:class", "ophyd_async.epics.adaravis._aravis.AravisDetector"),
     ("py:class", "AravisDetector"),
     ("py:class", "NoneType"),
     ("py:class", "'str'"),
diff --git a/pyproject.toml b/pyproject.toml
index c51960f..5aa10ee 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -13,7 +13,7 @@ classifiers = [
 description = "Bluesky playground environment for Diamond's htss rigs"
 dependencies = [
     "ophyd",
-    "ophyd-async==0.4.0",
+    "ophyd-async>=0.8.0",
     "databroker[client]==2.0.0b17",
     "matplotlib<3.8",
     "dls-dodal",
@@ -50,6 +50,8 @@ dev = [
     "sphinx-design",
     "tox-direct",
     "types-mock",
+    "blueapi>=0.7.0",
+    "pytest-env",
 ]
 
 [project.scripts]
@@ -78,6 +80,9 @@ addopts = """
 filterwarnings = "error"
 # Doctest python code in docs, python code in src docstrings, test functions in tests
 testpaths = "docs tests"
+env= [
+    "BEAMLINE=",
+]
 
 [tool.coverage.run]
 data_file = "/tmp/htss_rig_bluesky.coverage"
@@ -93,7 +98,7 @@ legacy_tox_ini = """
 [tox]
 skipsdist=True
 
-[testenv:{pre-commit,type-checking,tests,docs}]
+[testenv:{pre-commit,type-checking,tests,docs,system-test}]
 # Don't create a virtualenv for the command, requires tox-direct plugin
 direct = True
 passenv = *
@@ -106,8 +111,9 @@ allowlist_externals =
 commands =
     pre-commit: pre-commit run --all-files --show-diff-on-failure {posargs}
     type-checking: mypy src tests {posargs}
-    tests: pytest --cov=htss_rig_bluesky --cov-report term --cov-report xml:cov.xml {posargs}
+    tests: pytest tests/unit_tests --cov=htss_rig_bluesky --cov-report term --cov-report xml:cov.xml {posargs}
     docs: sphinx-{posargs:build -EW --keep-going} -T docs build/html
+    system-test: pytest tests/system_tests {posargs}
 """
 
 [tool.ruff]
diff --git a/src/htss_rig_bluesky/devices.py b/src/htss_rig_bluesky/devices.py
index f40b232..b403586 100644
--- a/src/htss_rig_bluesky/devices.py
+++ b/src/htss_rig_bluesky/devices.py
@@ -8,9 +8,9 @@
     Device,
     StaticPathProvider,
 )
-from ophyd_async.epics.areadetector.aravis import AravisDetector
-from ophyd_async.epics.motion import Motor
-from ophyd_async.epics.signal import epics_signal_rw
+from ophyd_async.epics.adaravis import AravisDetector
+from ophyd_async.epics.core import epics_signal_rw
+from ophyd_async.epics.motor import Motor
 
 from .names import pv_prefix
 
diff --git a/src/htss_rig_bluesky/plans/calibration.py b/src/htss_rig_bluesky/plans/calibration.py
index 8d57388..b420432 100644
--- a/src/htss_rig_bluesky/plans/calibration.py
+++ b/src/htss_rig_bluesky/plans/calibration.py
@@ -1,3 +1,4 @@
+import asyncio
 from collections.abc import Generator
 
 import bluesky.plan_stubs as bps
@@ -41,11 +42,14 @@ def scan_center(
     Yields:
         Plan
     """
-
-    x_range = abs(sample.x.high_limit_travel - sample.x.low_limit_travel)
+    limits = asyncio.gather(
+        sample.x.high_limit_travel.get_value(), sample.x.low_limit_travel.get_value()
+    )
+    high, low = limits.result()
+    x_range = abs(high - low)
     limit_margin = x_range * 0.01
-    min_x = min_x or sample.x.low_limit_travel + limit_margin
-    max_x = max_x or sample.x.high_limit_travel - limit_margin
+    min_x = min_x or low + limit_margin
+    max_x = max_x or high - limit_margin
 
     yield from bps.mv(
         det.drv.num_images,
diff --git a/src/htss_rig_bluesky/plans/detector.py b/src/htss_rig_bluesky/plans/detector.py
index f7f10ff..b21e9af 100644
--- a/src/htss_rig_bluesky/plans/detector.py
+++ b/src/htss_rig_bluesky/plans/detector.py
@@ -2,7 +2,7 @@
 from dataclasses import dataclass
 
 import bluesky.plan_stubs as bps
-from ophyd_async.epics.areadetector.aravis import AravisDetector
+from ophyd_async.epics.adaravis import AravisDetector
 
 
 @dataclass
diff --git a/tests/system_tests/__init__.py b/tests/system_tests/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/system_tests/conftest.py b/tests/system_tests/conftest.py
new file mode 100644
index 0000000..cf2ebd6
--- /dev/null
+++ b/tests/system_tests/conftest.py
@@ -0,0 +1,52 @@
+import pytest
+from blueapi.client.client import BlueapiClient
+from blueapi.config import ApplicationConfig, RestConfig, StompConfig
+from blueapi.worker.task import Task
+from bluesky_stomp.models import BasicAuthentication
+
+from htss_rig_bluesky.names import BEAMLINE
+
+
+@pytest.fixture
+def task_definition() -> dict[str, Task]:
+    return {
+        "step_scan_plan": Task(
+            name="step_scan_plan",
+            params={"detectors": "det", "motor": "sample_stage.theta"},
+        ),
+        "fly_and_collect_plan": Task(
+            name="fly_and_collect_plan",
+            params={"panda": "panda", "diff": "det"},
+        ),
+        "log_scan_plan": Task(
+            name="log_scan_plan",
+            params={"detectors": "det", "motor": "sample_stage.x"},
+        ),
+    }
+
+
+@pytest.fixture
+def config() -> ApplicationConfig:
+    if BEAMLINE == "p46":
+        return ApplicationConfig(
+            stomp=StompConfig(
+                host="172.23.168.198",
+                auth=BasicAuthentication(username="guest", password="guest"),  # type: ignore
+            ),
+            api=RestConfig(
+                host="p46-blueapi.diamond.ac.uk", port=443, protocol="https"
+            ),
+        )
+    else:
+        return ApplicationConfig(
+            stomp=StompConfig(
+                host="localhost",
+                auth=BasicAuthentication(username="guest", password="guest"),  # type: ignore
+            )
+        )
+
+
+# This client will use authentication if a valid cached token is found
+@pytest.fixture
+def client(config: ApplicationConfig) -> BlueapiClient:
+    return BlueapiClient.from_config(config=config)
diff --git a/tests/system_tests/test_plans.py b/tests/system_tests/test_plans.py
new file mode 100644
index 0000000..af582eb
--- /dev/null
+++ b/tests/system_tests/test_plans.py
@@ -0,0 +1,75 @@
+import pytest
+from blueapi.client.client import BlueapiClient
+from blueapi.client.event_bus import AnyEvent
+from blueapi.core.bluesky_types import DataEvent
+from blueapi.worker.event import TaskStatus, WorkerEvent, WorkerState
+from blueapi.worker.task import Task
+
+# Please export BEAMLINE=pXX before running the tests or add it in pyproject.toml
+
+
+def _check_all_events(all_events: list[AnyEvent]):
+    assert (
+        isinstance(all_events[0], WorkerEvent) and all_events[0].task_status is not None
+    )
+    task_id = all_events[0].task_status.task_id
+    # First event is WorkerEvent
+    assert all_events[0] == WorkerEvent(
+        state=WorkerState.RUNNING,
+        task_status=TaskStatus(
+            task_id=task_id,
+            task_complete=False,
+            task_failed=False,
+        ),
+    )
+
+    assert all(
+        isinstance(event, DataEvent) for event in all_events[1:-2]
+    ), "Middle elements must be DataEvents."
+
+    # Last 2 events are WorkerEvent
+    assert all_events[-2:] == [
+        WorkerEvent(
+            state=WorkerState.IDLE,
+            task_status=TaskStatus(
+                task_id=task_id,
+                task_complete=False,
+                task_failed=False,
+            ),
+        ),
+        WorkerEvent(
+            state=WorkerState.IDLE,
+            task_status=TaskStatus(
+                task_id=task_id,
+                task_complete=True,
+                task_failed=False,
+            ),
+        ),
+    ]
+
+
+@pytest.mark.parametrize("device", ["sample_stage", "panda", "det"])
+def test_device_present(client: BlueapiClient, device: str):
+    assert client.get_device(device), f"{device} is not available"
+
+
+@pytest.mark.parametrize(
+    "plan", ["step_scan_plan", "fly_and_collect_plan", "log_scan_plan"]
+)
+def test_spec_scan_task(
+    client: BlueapiClient,
+    task_definition: dict[str, Task],
+    plan: str,
+):
+    assert client.get_plan(plan), f"In {plan} is available"
+
+    all_events: list[AnyEvent] = []
+
+    def on_event(event: AnyEvent):
+        all_events.append(event)
+
+    client.run_task(task_definition[plan], on_event=on_event)
+
+    _check_all_events(all_events)
+
+    assert client.get_state() is WorkerState.IDLE
diff --git a/tests/conftest.py b/tests/unit_tests/conftest.py
similarity index 100%
rename from tests/conftest.py
rename to tests/unit_tests/conftest.py
diff --git a/tests/test_boilerplate_removed.py b/tests/unit_tests/test_boilerplate_removed.py
similarity index 100%
rename from tests/test_boilerplate_removed.py
rename to tests/unit_tests/test_boilerplate_removed.py
diff --git a/tests/test_cli.py b/tests/unit_tests/test_cli.py
similarity index 100%
rename from tests/test_cli.py
rename to tests/unit_tests/test_cli.py
diff --git a/tests/test_imports.py b/tests/unit_tests/test_imports.py
similarity index 100%
rename from tests/test_imports.py
rename to tests/unit_tests/test_imports.py