diff --git a/tensorflow_datasets/datasets/smart_buildings/controller_reader.py b/tensorflow_datasets/datasets/smart_buildings/controller_reader.py
index fba69d9b53b..119da9b6c38 100644
--- a/tensorflow_datasets/datasets/smart_buildings/controller_reader.py
+++ b/tensorflow_datasets/datasets/smart_buildings/controller_reader.py
@@ -22,6 +22,7 @@
 from typing import Callable, Mapping, Sequence, TypeVar, Union
 
 from absl import logging
+from etils import epath
 import pandas as pd
 from tensorflow_datasets.datasets.smart_buildings import constants
 from tensorflow_datasets.datasets.smart_buildings import reader_lib
@@ -45,8 +46,8 @@ class ProtoReader(reader_lib.BaseReader):
     input_dir: directory path where the files are located
   """
 
-  def __init__(self, input_dir):
-    self._input_dir = input_dir
+  def __init__(self, input_dir: epath.PathLike):
+    self._input_dir = epath.Path(input_dir)
     logging.info('Reader lib input directory %s', self._input_dir)
 
   def read_observation_responses(
@@ -97,7 +98,7 @@ def read_reward_responses(  # pytype: disable=signature-mismatch  # overriding-r
 
   def read_zone_infos(self) -> Sequence[smart_control_building_pb2.ZoneInfo]:
     """Reads the zone infos for the Building from .pbtxt."""
-    filename = os.path.join(self._input_dir, constants.ZONE_INFO_PREFIX)
+    filename = self._input_dir / constants.ZONE_INFO_PREFIX
     return self._read_streamed_protos(
         filename, smart_control_building_pb2.ZoneInfo.FromString
     )
@@ -107,7 +108,7 @@ def read_device_infos(
   ) -> Sequence[smart_control_building_pb2.DeviceInfo]:
     """Reads the device infos for the Building."""
 
-    filename = os.path.join(self._input_dir, constants.DEVICE_INFO_PREFIX)
+    filename = self._input_dir / constants.DEVICE_INFO_PREFIX
     return self._read_streamed_protos(
         filename, smart_control_building_pb2.DeviceInfo.FromString
     )
@@ -141,28 +142,26 @@ def _read_messages(
       messages.extend(file_messages)
     return messages
 
-  def _read_shards(self, input_dir: str, file_prefix: str) -> Sequence[str]:
+  def _read_shards(
+      self, input_dir: epath.Path, file_prefix: str
+  ) -> Sequence[epath.Path]:
     """Returns full paths in input_dir of files starting with file_prefix."""
-
-    shards = [
-        os.path.join(input_dir, f)
-        for f in os.listdir(input_dir)
-        if f.startswith(file_prefix)
-    ]
-    return shards
+    return list(epath.Path(input_dir).glob(f'{file_prefix}*'))
 
   def _select_shards(
       self,
       start_time: pd.Timestamp,
       end_time: pd.Timestamp,
-      shards: Sequence[str],
-  ) -> Sequence[str]:
+      shards: Sequence[epath.Path],
+  ) -> Sequence[epath.Path]:
     """Returns the shards that fall inside the start and end times."""
 
-    def _read_timestamp(filepath: str) -> pd.Timestamp:
+    def _read_timestamp(filepath: epath.Path) -> pd.Timestamp:
       """Reads the timestamp from the filepath."""
       assert filepath
-      ts = pd.Timestamp(re.findall(r'\d{4}\.\d{2}\.\d{2}\.\d{2}', filepath)[-1])
+      ts = pd.Timestamp(
+          re.findall(r'\d{4}\.\d{2}\.\d{2}\.\d{2}', os.fspath(filepath))[-1]
+      )
       return ts
 
     def _between(
@@ -179,13 +178,13 @@ def _between(
 
   def _read_streamed_protos(
       self,
-      full_path: str,
+      full_path: epath.Path,
       from_string_func: Callable[[Union[bytearray, bytes, memoryview]], T],
   ) -> Sequence[T]:
     """Reads a proto which has byte size preceding the message."""
 
     messages = []
-    with open(full_path, 'rb') as f:
+    with full_path.open('rb') as f:
       while True:
         # Read size as a varint
         size_bytes = f.read(4)
@@ -260,7 +259,7 @@ def get_episode_data(working_dir: str) -> pd.DataFrame:
   Returns:
     A dataframe with episode label, timestamps, number of updates.
   """
-  episode_dirs = os.listdir(working_dir)
+  episode_dirs = list(epath.Path(working_dir).iterdir())
   date_extractor = operator.itemgetter(slice(-13, None))
 
   execution_times = pd.to_datetime(