The Driver Service (EgodriverService) implements the autonomous driving policy. It receives sensor observations and produces planned trajectories for the ego vehicle.
Service Definition service EgodriverService { rpc start_session ( DriveSessionRequest ) returns ( common . SessionRequestStatus ); rpc close_session ( DriveSessionCloseRequest ) returns ( common . Empty ); rpc submit_image_observation ( RolloutCameraImage ) returns ( common . Empty ); rpc submit_egomotion_observation ( RolloutEgoTrajectory ) returns ( common . Empty ); rpc submit_route ( RouteRequest ) returns ( common . Empty ); rpc submit_recording_ground_truth ( GroundTruthRequest ) returns ( common . Empty ); rpc drive ( DriveRequest ) returns ( DriveResponse ); rpc get_version ( common . Empty ) returns ( common . VersionId ); rpc shut_down ( common . Empty ) returns ( common . Empty ); }
Methods start_session Initialize a driving session with vehicle and scene configuration.
Unique identifier for this driving session
Random seed for reproducible behavior
Vehicle specification Show RolloutSpec.VehicleDefinition
List of cameras available on the vehicle with intrinsics and extrinsics
Debug information (not present in benchmark runs) Scene identifier for debugging
submit_image_observation Provide camera images to the driver for perception.
Frame capture start timestamp (microseconds)
Frame capture end timestamp (for rolling shutter)
Encoded image data (PNG/JPEG)
Camera identifier (e.g., “camera_front_wide_120fov”)
submit_egomotion_observation Provide estimated ego vehicle trajectory (localization output).
Estimated ego rig pose as active transform local→rig_est. This can diverge from the true local→rig transform when error models are enabled.
Estimated velocities and accelerations at current timestamp in rig frame.
May include noise when error models are active.
submit_route Provide high-level route waypoints to the driver.
Timestamp at which the route is valid
Waypoints expressed in rig frame at the provided timestamp
submit_recording_ground_truth Provide ground truth trajectory from recording (for testing/validation).
Timestamp at which the ground truth is valid
Ground-truth trajectory from recording (actual path followed).
Each pose is expressed in rig frame at the provided timestamp.
drive Request a planned trajectory from the driver.
Current simulation time (microseconds)
Time for which trajectory is requested
Optional arbitrary data from renderer for advanced use cases
Planned trajectory as active transform local→rig_est produced by the driver
Binary debug data with custom serialization
Alternative trajectories considered by planner
close_session Release resources for a session.
Session identifier to close
Usage Example From alpasim_runtime/services/driver_service.py:
from alpasim_grpc.v0 import egodriver_pb2, egodriver_pb2_grpc import grpc # Connect to driver service channel = grpc.insecure_channel( 'localhost:50051' ) driver_stub = egodriver_pb2_grpc.EgodriverServiceStub(channel) # Start session session_request = egodriver_pb2.DriveSessionRequest( session_uuid = "my-session" , random_seed = 42 , rollout_spec = egodriver_pb2.DriveSessionRequest.RolloutSpec( vehicle = egodriver_pb2.DriveSessionRequest.RolloutSpec.VehicleDefinition( available_cameras = camera_specs ) ) ) driver_stub.start_session(session_request) # Submit observations driver_stub.submit_image_observation( egodriver_pb2.RolloutCameraImage( session_uuid = "my-session" , camera_image = egodriver_pb2.RolloutCameraImage.CameraImage( frame_start_us = t_start, frame_end_us = t_end, image_bytes = image_data, logical_id = "camera_front_wide_120fov" ) ) ) driver_stub.submit_egomotion_observation( egodriver_pb2.RolloutEgoTrajectory( session_uuid = "my-session" , trajectory = ego_trajectory, dynamic_state = ego_state ) ) # Request driving plan response = driver_stub.drive( egodriver_pb2.DriveRequest( session_uuid = "my-session" , time_now_us = current_time, time_query_us = query_time ) ) planned_trajectory = response.trajectory print ( f "Planned { len (planned_trajectory.poses) } poses" ) # Clean up driver_stub.close_session( egodriver_pb2.DriveSessionCloseRequest( session_uuid = "my-session" ) )
Data Flow The driver service follows this typical workflow:
Session Initialization : Configure vehicle and sceneObservation Loop :
Receive camera images via submit_image_observation
Receive localization via submit_egomotion_observation
Receive route via submit_route
Planning : Generate trajectory with driveSession Cleanup : Close session when done