Integrating a Robot

Integrating a Robot#

How to run open_apero_node on a robot of your own: the nodes to keep, what the robot provides and what to configure. ROS Node, Launch Files and Messages describes the node.

Start from a copy of run_experiment_sim.launch: keep voxblox_node, open_apero_node and mission_coordinator_node, and drop the HSSD scene-file loads, habitat_renderer_node, dummy_controller_node and, unless an RGB-D camera publishes exactly synchronized camera info, RGB and 32-bit float depth, rgbd_to_pointcloud_node. Give the coordinator <param name="wait_for_simulation" value="false"/>, or it waits for Habitat and ends the launch after 30 s.

What the Robot Provides#

  • Commands: execute each /command/trajectory (open_apero_msgs/RobotTrajectory, in map_frame, a launch argument, world by default). Its states hold the poses of base_frame and camera_frame (x forward, z up) and a time_from_start, robot.time_step (0.2 s) apart; a planning command has at most replan_num_steps + 1 (6 shipped) of them. The next command comes once this one is done.

  • Pose with use_tf: false (shipped, for a lockstep simulator): publish /robot_state (open_apero_msgs/RobotState, the same two poses) while executing a command and, until the first one, the initial pose, which dummy_controller_node repeats every 0.1 s. Each message counts as the outcome of the latest command.

  • Pose with use_tf: true (for an asynchronous robot): publish TF map_frame → base_frame and map_frame → camera_frame, and after each command one /command/result (open_apero_msgs/CommandResult) with its command_sequence, header.frame_id = map_frame and the final state. TF alone never completes a command.

  • Point clouds: publish /open_apero_camera/point_cloud (sensor_msgs/PointCloud2 in camera_optical_frame) and /camera_transform (geometry_msgs/TransformStamped, that frame in map_frame) with identical stamps, or set voxblox_node’s use_tf_transforms to true and remap its pointcloud input to the robot’s cloud. Give pixels with no return as points beyond ray_length, like Habitat’s far_depth_sentinel: only these clear free space along rays that hit nothing.

A command is done once the sensor pose is within target_position_threshold (0.10 m) and target_angle_threshold (0.20 rad) of its last state and, with wait_for_map_updated: true, a cloud taken there in the optical frame of camera_frame is mapped (The Settle Gate). For a cloud in another frame, such as a LiDAR’s, set wait_for_map_updated: false; for a robot that cannot track within the thresholds, raise them. Poses with a non-unit quaternion are dropped. Of a RobotState on /robot_state or /command/result, the planner keeps the camera pose, roll zeroed, and re-derives the base from it through the robot model’s sensor mount.

Adapt the Configuration#

  • robot in config/planners/_common.yaml: velocity limits, collision spheres, sample_z: false for a ground robot and, with use_tf: false, the sensor mount (Robot Models).

  • /map_bounding_volume (where the robot may plan) and /target_bounding_volume (where gain counts), boxes in map_frame (Configuration Files).

  • sensor_model in config/tasks/_common.yaml: the camera’s resolution and focal length. A LiDAR takes IterativeRayCasterLidar, which is spherical: give it resolution_x/_y (rays) and field_of_view_x/_y [deg] instead of focal_length and observation_boundary, and drop the launch’s sensor_model/observation_boundary param, or they are reported unread (Sensor Models). The launch arguments ray_length and observation_boundary set the camera’s range and voxblox_node’s; set them to what the robot’s map integrates with.

Start and Stop#

The coordinator creates the run directory, sets the planner’s ~run_dir and runs it until time_limit minutes pass or a message arrives on /open_apero/collision (std_msgs/String), the hook for a bumper or safety node. To start the planner by hand, give it an existing, empty run directory first, which it requires with log_data:=true (default):

mkdir -p ~/open_apero_data/my_robot/run_01
rosparam set /open_apero_node/run_dir ~/open_apero_data/my_robot/run_01
rosservice call /open_apero_node/toggle_running "data: true"   # false stops, writes the summary

The open_apero_node reference lists every topic, service and parameter. A platform without ROS derives its own agent from Agent and implements sendMovement() and publishVisualization() (The Loop).