Fetch Robotics is committed to providing an exceptional out-of-the-box experience for new robot owners. One of the ways in which we create a great experience is by using standard ROS interfaces. This means that code you might have written for other robots in the past should be easily portable to your new robot.
Whenever possible, we have conformed to the ROS Enhancement Proposals (REPs). These documents provide the foundation of standard ROS interfaces. In addition to REP-compatible interfaces, we have adopted a number of the community-accepted standard interfaces, such as those provided by the control_msgs package.
Arm and Torso¶
The arm and torso of the robot are controlled by control_msgs/FollowJointTrajectory actions. This action interface is the output of packages such as MoveIt, and allows the arm to execute a pre-defined trajectory. Three interfaces are provided:
- arm_controller/follow_joint_trajectory to control just the seven joints of the arm.
- arm_with_torso_controller/follow_joint_trajectory to control the seven joints of the arm plus the torso.
- torso_controller/follow_joint_trajectory to control just the torso.
Only one controller is allowed to control a joint at a time.
Fetch is tuned to work with smooth paths. The tuning performs best when paths are properly smoothed and timed, such as those generated from MoveIt! Fetch has a number of mechanisms in place to prevent damage from improper trajectories. Improperly constructed paths may cause motor or breaker shutdown.
In addition to the trajectory controllers, the arm is always running a gravity compensation controller.
Support for mobile bases is quite standard and robust in ROS, however it is one of the older interfaces. As such, it is one of the few interfaces which is not action-based.
The mobile base subscribes to base_controller/command, and accepts a geometry_msgs/Twist message.
Only two fields are used in the message:
linear.xspecifies the robot’s forward velocity
angular.zspecifies the robot’s turning velocity
User applications will typically not connect directly to base_controller/command, but rather to cmd_vel. A multiplexer is always running between cmd_vel/teleop and cmd_vel. Whenever the deadman on the robot controller is held, cmd_vel/teleop will override cmd_vel. The advantage of having your application publish to cmd_vel rather than directly to base_controller/command is that you can override bad commands by simply pressing the deadman on the robot controller.
The base controller implements a speed reduction when in the proximity of obstacles. This will not entirely stop the robot if it is about to hit something, but will prevent full speed collisions.
The head exposes two potential interfaces. The first is an ActionServer available on head_controller/follow_joint_trajectory which follows a joint trajectory as with the arm and torso (described above).
The second interface is unique to the head, and allows the user to easily point the
head (and head sensors) at a point of interest. This action-based interface is
available on head_controller/point_head. It is of type
Although the interface currently does not support any of the
fields, it points the head_tilt_link (which is very near the camera optical
axis) towards the target point to achieve a similar effect. A
max_velocity can also be specified.
gripper_controller/gripper_action exposes a
ActionServer. The gripper command takes in
effort as parameters. Generally,
the gripper is commanded to a fully closed or fully opened position, so
effort is used to limit the maximum effort. As the gripper never fully reaches
the closed position, the grasp strength will be determined by the maximum
Head Camera Interface¶
The head camera exposes several topics of interest:
- head_camera/depth_registered/points is a sensor_msgs/PointCloud2 which has both 3d and color data. It is published at VGA resolution (640x480) at 15Hz.
- head_camera/depth_downsampled/points is a sensor_msgs/PointCloud2 which has only 3d data. It is published at QQVGA (160x120) resolution at 15Hz and is intended primarily for use in navigation/moveit for obstacle avoidance.
- head_camera/depth/image_raw is a sensor_msgs/Image. This is unit16 depth image (2D) in mm . It is published at VGA resolution (640x480) at 15Hz.
- head_camera/depth/image is a sensor_msgs/Image. This is float depth image (2D) in m. It is published at VGA resolution (640x480) at 15Hz.
- head_camera/rgb/image_raw is a sensor_msgs/Image. This is just the 2d color data. It is published at VGA resolution (640x480) at 15Hz.
base_scan is a sensor_msgs/LaserScan message published at 15Hz.
Note: the raw laser information as reported by the laser hardware is published to base_scan_raw. The information published to base_scan is filtered to remove shadow points.
imu is a sensor_msgs/Imu message published at 100Hz. This message contains the linear acceleration and rotational velocities as measured by the IMU located in the base of the robot.
On Fetch robots, the gripper IMU publishes to gripper_imu. This is also a sensor_msgs/Imu message published at 100Hz.
The IMUs are not present in the simulated robot.
There are 3 breakers governing power on the fetch. One each for the arm, gripper and base. If the motors are commanded to perform beyond their limits they will shut down for safety. To reset them you will either need to toggle the e-stop or use the following service calls:
- $ rosservice call /arm_breaker false && rosservice call /arm_breaker true
- $ rosservice call /base_breaker false && rosservice call /base_breaker true
- $ rosservice call /gripper_breaker false && rosservice call /gripper_breaker true