rosbridge was improved.set_jacobian_inverse_damping in the Python PathIKOptions API, allowing users to tune the Jacobian inverse damping factor for Path IK solving.project3dToPixel segfaulting or returning incorrect coordinates on Jetson ARM64 due to an OpenCV ABI mismatch between the pre-built ros-humble-image-geometry package and NVIDIA's JetPack OpenCV. All objectives using GetMasks3DFromMasks2D or FindMaskedObjects now work correctly on both amd64 and ARM64.moveit_pro_acm_updater tool to handle the case where the output SRDF already exists by adding an -f/--force flag and merging new ACM entries into an existing SRDF instead of failing or requiring manual cleanup.
ROS Jazzy builds are now available, alongside our existing ROS Humble build! Support for ROS Jazzy is still experimental, but all you need to do to try it out is set the variable MOVEIT_ROS_DISTRO to jazzy in your ~/.config/moveit_pro/moveit_pro_config.9.yaml file.
SAM3-based multimodal image segmentation is now available through the new GetMasks2DFromExemplar Behavior. This capability enables flexible object detection using combinations of text prompts, exemplar images, and bounding boxes, allowing developers to build robust exemplar-based perception pipelines. To help users get started, the lab_sim environment now includes three new Objectives demonstrating exemplar-driven segmentation workflows: ML Segment Bottles from File, ML Find Objects on Table, and ML Find Bottles on Table from Image Exemplar.
The CPU-only ONNX (Open Neural Network Exchange) inference now uses all available cores and enables memory pooling, providing a significant speedup (4-8x) on machines without a GPU. Previously, threading and memory settings were tuned for GPU execution and left CPU inference single-threaded. Also, when no GPU is available and ONNX Runtime falls back to CPU inference, a warning is now surfaced in the UI log panel. Previously this was only printed to the console, so users had no visibility into why ML inference was slower than expected.
You can now add custom dashboards, data displays, and web content as custom view panes in the canvas. You can add external URLs or local HTML files, and optionally enable ROS Bridge for each pane to allow content to interact with ROS topics. Access the management modal by clicking "Add Custom Pane" in any view pane's dropdown menu to add, edit, or remove custom view panes. An example Joint States Monitor pane (joint-states-monitor-example.html) is included as a reference for building custom panes with live ROS data.
See the Creating Custom View Panes in the MoveIt Pro UI for more details.
The navigation bar now shows the name(s) of currently running Behavior Tree leaf nodes during Objective execution instead of the static "Executing..." text.
Hovering over teleop buttons now shows a transparent ghost robot previewing the target position. This works for Pose teleop D-pad buttons, Joint teleop +/- buttons, and gripper open/close buttons (joints at their limit positions). The ghost hides during button press and reappears on the next hover.
The UI has been streamlined by consolidating navigation controls into the View dropdown menu, where navigation costmaps (Occupancy Grid, Global Costmap, and Local Costmap) now appear alongside Display options and Markers in a single visibility toggle. Additionally, the View Pane menu has been redesigned to show all available panes with checkboxes indicating the active one, making it easier to see and switch between views without hiding the currently selected pane.
The Keep-Out Zone modal has been replaced with a dedicated Planning Scene Editor sidebar, making it easier to create and edit environment objects directly within the UI. The new panel supports adjustable Height, Width, and Length for non-cube shapes, precise position and rotation controls (X/Y/Z and Roll/Pitch/Yaw), object deletion, and quick access to Load Scene and Export Scene tools.
A new coordinate frame indicator in the 3D viewport shows labeled XYZ axes that rotate with the camera, helping you identify the world coordinate frame at a glance. The indicator can be toggled from the View menu.
ListControllers is a new Behavior to retrieve controllers based on their type and activation state.SetMujocoState is a new Behavior to set the joint state of the robot inside MuJoCo simulation.CreateBoundingBox2D is a new Behavior that constructs a BoundingBox2D message from individual coordinate and size ports, suitable for wiring into Behaviors that accept box prompts.CreateBoundingBoxes2D is a new Behavior that converts a flat vector of doubles into a vector of BoundingBox2D messages, where each group of four values (x, y, width, height) specifies a single box.CreateBoundingBoxFromOffset is a new Behavior that generates a BoundingBox2D covering an exemplar image's dimensions with optional padding, ready to be passed to GetMasks2DFromExemplar.PublishBoundingBoxes2D is a new Behavior that publishes a visualization of labeled bounding boxes overlaid on an image to a ROS topic.OverridePoseOrientation is a new Behavior that replaces the orientation of a pose with a fixed quaternion, useful for enforcing gripper-down approach orientations.GetCenterMostAprilTag is a new Behavior that selects the AprilTag detection closest to the camera image center from a set of detections.ClearAllVisualMarkers is a new Behavior that deletes all visualization markers from the /visual_markers topic.VisualizeLine is a new Behavior that draws a line marker between two poses, useful for visualizing spatial relationships such as detection-to-grasp offsets.SwitchControllers now detects control chains and automatically activates or deactivates controllers in a chain if automatic_activation or automatic_deactivation are set to true.PlanMTCTask (MoveIt Task Constructor) now has a suppress_failure_message port that silently returns failure instead of publishing a UI alert when MTC planning fails. This is useful when planning failures are expected and handled by retry decorators.ResetMujocoSim disables all active command controllers before resetting MuJoCo, and re-enables them after.TransformPose can now optionally visualize the resulting pose as a coordinate-frame marker.VisualizePath now supports customizable line color (line_color_rgb port, RGB 0-255) and line width (line_width port, in meters). Defaults match the previous green color and 1 cm width, with the line opacity increased from 0.8 to 1.0 for improved visibility.MoveGripperAction provides improved timeout error messages with current port values and actionable troubleshooting steps.PlanMTCTask error messages now include the task name, indent each failing stage for readability, and include a legend explaining the (solutions found / failures) notation.SetupMTCFixedJointState now has an ignore_collisions port that allows users to skip collision checking in the FixedState stage. This is useful when the initial planning scene is expected to be in collision and the ACM rules are not yet set up, allowing downstream stages with the correct ACM to handle collisions instead of failing immediately.lab_sim_behaviors package for lab_sim with a ComputeTrayPlacePositionsUsingAprilTags custom Behavior that computes grid-based place positions relative to a tray AprilTag for bottle placement workflows.SetMujocoState in lab_sim demonstrating how to set the robot's joint state inside the MuJoCo simulation.lab_sim there are three new Objectives to demonstrate how to use exemplars with GetMasks2DFromExemplar.
ML Segment Bottles from File loads an image from file of a laboratory table and uses text prompting to find bottles. Useful for troubleshooting and tuning parameters and prompts.ML Find Objects on Table uses text prompting to find various objects on the MuJoCo lab table.ML Find Bottles on Table from Image Exemplar uses an image exemplar of a square pill bottle to find simulated pill bottles in MuJoCo scene. Combine image exemplar with text prompts for more performance.set_jacobian_inverse_damping in the Python PathIKOptions API, allowing users to tune the Jacobian inverse damping factor for Path IK solving.SetupMTCFixedJointState checked for collisions using its own isolated planning scene, ignoring ACM (Allowed Collision Matrix) modifications from upstream MTC stages like SetupMTCIgnoreCollisionsBetweenObjects. A new ignore_collisions port (default: true) defers collision checking to downstream stages that have the correct ACM. To restore the previous Behavior, set ignore_collisions to false.ModifyPlanningScene stage. This stage is used to modify the ACM and add/remove collision objects, not to plan motion. The unscoped collision check caused false failures when unrelated objects in the scene were colliding (e.g., an octomap overlapping a static collision object).MergePointClouds where NaN (Not a Number) values in the point clouds could cause ICP (Iterative Closest Point) alignment to fail and throw an exception. The Behavior now removes NaN values from the point clouds before alignment to ensure robust operation.hangar_sim planning scene. The access stand (collision_Cube_002.001) was not in the URDF, and the cart (collision_Cube) position did not match its Mujoco world location, causing the robot to plan through both objects.pose port value for VisualizePose to use {pose_stamped} (previously {stamped_pose}) to match MoveIt Pro conventions.This is a very exciting major release, full of new features! 🎁
MoveIt Pro 9.0 introduces a new Joint Jog teleoperation mode that allows users to directly control individual robot joints through the Web UI. This new implementation replaces the previous MoveIt Servo-based joint jogging with a more robust and performant solution.
Key improvements over the previous Servo-based implementation:
joint_jog.yaml) that maps planning groups to their controllers, eliminating the need for complex MoveIt Servo parameter tuning.The Joint Jog Behavior performs continuous collision checking to ensure safe operation, with user-configurable link padding and stop safety factors. Users can also disable collision checking when needed for specific applications.
See the 9.0 User Workspace Migration Guide for instructions on updating your robot configuration packages to use the new Joint Jog feature, as the previous MoveIt Servo-based implementation is no longer available.
One of the recent improvements to MoveIt Pro is a set of Behaviors that enable scanning point clouds of the real-world and automatically generating rasterized Cartesian tool paths across the contours. This is ideal for spraying, washing, sanding, grinding, and more. You can see an example of this working in the Move along Contour Objective in the hangar_sim world and Plan Path Along Surface Objective in factory_sim. New Behaviors include:
GetContourFromPointCloudSlice - creates a planar contour that is offset around a point cloud slice.MergePointClouds - exposes an option to align the point clouds using ICP when merging them.CropPosesInBox - filter a vector of poses, such as a Cartesian path, based on a bounding box.
New Behaviors were added to enable inspecting and filtering bounding boxes and areas of 2D masks.
Check out FilterMasks2DByArea, FilterMasks2DByBoundingBox, and GetMask2DProperties for more information.
Useful for pruning image masks produced by ML Behaviors such as automasking.
The new 9.0 UI reduces mouse travel for common workflows, removes confusion around editing while in Build mode, and eliminates wasted space caused by duplicate header bars. The refreshed layout delivers a cleaner, more modern look that improves usability today while making the product more visually appealing. A key change is we have removed the main Build / Run buttons, unifying this concept into a single sidebar, but many other enhancements are explained below.
This editing mode allows nodes to be visually snapped to insertion points between siblings when dragging, making it easier to reorganize the tree structure. The implementation includes visual drop zones, grid-based layout for consistent node sizing, and a toggle in the UI to enable/disable the feature depending on your preference.
You can now search Behavior Tree nodes and ports by name in the editor, navigate between matches, and see the current match highlighted directly on the canvas.
Users can now filter alert messages in the alert history drawer by their source: Runtime (ROS logs and Objective execution events) or Editor (Build mode and editor operations). This makes it easier to focus on the messages that matter for the current task—for example, isolating Runtime alerts while debugging a Behavior Tree, or viewing only Editor-related notices while configuring Objectives.
Users can now view the robot and scene from multiple perspectives simultaneously (the original 3D "Visualization" view plus 3 new orthographic views along the X, Y, and Z axes). The different views can be selected from the view dropdown menu.
Added two new pane view options for large monitors: 5-pane view and 6-pane view. These layouts provide additional flexibility for users who need to view multiple camera feeds and visualization panels simultaneously.
In the teleoperation interface, users can now adjust multiple joint sliders and approve all changes together before execution.
We have revamped our robot configuration guides to be better understood by AI Assistants and turn URDFs into fully functional robot configurations.
PathIKOptions::setTrackPathOrientationNullspaceTask().PathIKOptions::setMaxAllowedPathDeviation(). Available in the C++ and Python APIs.PathIKOptions::setNullspaceOverrideFunction(). This function receives the computed nullspace component and current joint values, and can modify or replace the nullspace component before it is applied. Available in the C++ and Python APIs.<std::allocator<void> > suffixes are removed, std::__cxx11::basic_string<...> is simplified to std::string, and default allocator template parameters in vectors are removed. This makes type names cleaner and easier to read, and allows users to enter simple type names that will match for type highlighting.SetupMTCMoveToPose, SetupMTCMoveToJointState, SetupMTCMoveToNamedState, and SetupMTCConnectWithTrajectory had been deprecated in version 7.1.0 and are now removed entirely in 9.0. Please use SetupMTCPlanToPose, SetupMTCPlanToJointState, SetupMTCPlanToNamedState, and SetupMTCConnectWithProRRT respectively.DetachURDF and RemoveURDFFromScene Behaviors now return success (no-op) when attempting to detach or remove a tool that doesn't exist, instead of returning an error.ExecuteFollowJointTrajectory and ExecuteTrajectoryWithAdmittance into a single ExecuteTrajectory Behavior that supports both standard FollowJointTrajectory action servers and Joint Trajectory Admittance Controller (JTAC) action servers via an execution_pipeline input port.GetContourFromPointCloudSlice: Renamed input port desired_pose_spacing to max_pose_spacing to better reflect its Behavior as an upper limit on waypoint spacing rather than a target spacing value. This is a breaking change that requires updating Objective files.GetImage, GetCameraInfo, GetPointCloud: Renamed input port timeout_sec to message_timeout_sec to better distinguish it from the publisher_timeout_sec port. This is a breaking change that requires updating Objective files.ExecuteMTCTask has been deprecated in favor of ExecuteMTCSolution and the new Execute MTC Solution subtree. The core Move to Pose and Request Teleoperation Objectives have been updated to use this new subtree internally. This will allow deprecating the old MoveIt 2 execution pipeline in favor of more transparent and direct communication with the robot controllers.The following previously deprecated Behaviors have been removed. If you have custom Objectives using any of these Behaviors, please refer to the 9.0 User Workspace Migration Guide.
ActivateControllersAddSubframeToObjectAddToolToSceneAppendOrientationConstraintAttachToolDetachToolGenerateObjectsInBoxGetGraspAndTwistSubframesGetMoveAlongArcSubframesInitializeMotionConstraintsLoadObjectiveParametersRemoveToolFromSceneServoTowardsPoseSetupMTCApproachGraspSetupMTCGenerateCuboidGraspsSetupMTCGenerateVacuumGraspsSetupMTCRetractFromGraspTeleoperateTwistTransformPoseFromYamlWriteCalibratedPoseToYamlfactory_sim there is a new Objective Automask Camera Iterate Masks which demonstrates using automask on binary camera images and visualizes each mask individually using breakpoints. This particular Objective is useful for tuning mask parameters and can be applied to improve ML workflows.GetMasks2DAutomask for ML-based automasking or automatic segmentation of imagesPathIKOptions::setTrackPathOrientationNullspaceTask().PathIKOptions::setMaxAllowedPathDeviation(). Available in the C++ and Python APIs.PathIKOptions::setNullspaceOverrideFunction(). This function receives the computed nullspace component and current joint values, and can modify or replace the nullspace component before it is applied. Available in the C++ and Python APIs.Pick 1 Pill Bottle and Pick All Pill Bottles in lab_sim, the text prompts have been changed to a more descriptive prompt, improving the performance and consistency of the image segmentation.joints and base_link/base_frame parameters. All example configurations which still use these parameters now accept planning_group_name instead.LogMessage Behavior not setting a source name, which caused the UI to incorrectly parse messages containing periods or colons.LoggerBase::publishFailureMessage, publishWarnMessage, and publishInfoMessage are no longer virtual. Tests mocking these methods should mock publishMessage(int32_t, const std::string&) instead.qos_profile field which is set to sensor_data to match the QoS settings of MuJoCo. This should also be the default for most cameras but if you are unable to view a camera stream in the web UI, you can verify that the web subscription QoS matches the camera publisher QoS with ros2 topic info -v /my/camera/topic and adjust the publisher QoS settings accordingly./studio_ui/ namespace have been renamed to use the /moveit_pro_ui/ prefix instead. For example, /studio_ui/store_joint_state → /moveit_pro_ui/store_joint_state, /studio_ui/motion_ended → /moveit_pro_ui/motion_ended, /studio_ui/move_joint → /moveit_pro_ui/move_joint, /studio_ui/move_end_effector → /moveit_pro_ui/move_end_effector, and /studio_ui/switch_primary_view → /moveit_pro_ui/switch_primary_view. Any custom nodes subscribing or publishing to these topics must be updated accordingly.STUDIO variable prefixes were made into MOVEIT prefixes instead and are now found in a new ~/.config/moveit_pro_config.9.yaml.Information on migrating a workspace from MoveIt Pro 8.x to 9.0 can be found in the 9.0 User Workspace Migration Guide.
MoveIt Pro 9.0 encourages users to switch to the Joint Trajectory Admittance Controller (JTAC) for improved safety and performance when executing trajectories.
The JTAC will respect the robot joint acceleration limits when it needs to stop a trajectory.
Make sure to set the stop_accelerations parameter in the controller configuration high enough that the robot can decelerate and stop quickly enough to meet your application's safety requirements.
ServiceClientBehaviorBase log statements by including the service name and timeout duration.rosbridge topic discovery QoS. Any applications on top of MoveIt Pro using rosbridge will also find improvements even if you are not using the UI.--verbose/-v).--include never flag in the update_acm script only working when combined with --verbose.update_acm overwriting the whole srdf file. Users are now prompted to use --force/-f to overwrite the original srdf.urdf_to_mjcf verb creating incorrect mjcf files due to similar mesh and geom names.
A new sub-product is now available (additional costs apply) that exposes the core MoveIt Pro planners, solvers, and controllers with low-level C++ and Python APIs. It does not include Behavior Trees nor much of the rest of MoveIt Pro's capabilities. Learn more..
Now when you edit a Behavior Tree XML file in your IDE / text editor of choice, the changes are automatically refreshed in the MoveIt Pro UI.
The urdf_to_mjcf makes it easier to convert between different robot model types - ROS' URDF and MuJoCo's MJCF format.
The new Behavior GetMasks2DAutomask uses the SAM2 model to automatically segment images using a grid of points as prompts. You can see an example of how this is utilized in the factory_sim robot configuration package.
A new Behavior ExecuteMTCSolution is available to execute a pre-computed MTC solution trajectory on a robot. This Behavior will replace the existing ExecuteMTCTask Behavior, as we advance towards a leaner execution pipeline.
ExecuteMTCSolution offers two controller interfaces: one for standard 'FollowJointTrajectory' controllers (e.g. ros2_control JTC), and another for the Joint Trajectory Admittance Controller (JTAC) included in MoveIt Pro.
This allows users to easily execute MTC solutions with or without admittance control, and simplifies integration with other vendor controllers different than the ros2_control JTC, but offering the same FollowJointTrajectory action interface.
tf2-tools is now automatically installed and available in the MoveIt Pro container.PlanToJointGoal) will now properly report when goal states are outside of the given orientation tolerance.AddVirtualObjectToPlanningScene Behavior where virtual objects were incorrectly set to allow collisions with all other objects by default.Added a toast message history list into a sidebar to view past errors, warnings, and info messages. Supports search and respects current log level settings.
Added a Visualize TF Objective to visualize any transform frame known to ROS' tf2 in the UI.
This can be used in your Behavior Trees to help with debugging and visualization of coordinate frames.
It requires core visualization Objectives to be enabled in the robot configuration package config.yaml file, for instance:
objectives:
...
objective_library_paths:
...
visualization_objectives:
package_name: "moveit_pro_objectives"
relative_path: "objectives/visualization"
Added a collapse/expand button for Sequence nodes, similar in functionality to Subtrees.
Added a visual difference in Behavior Trees to distinguish between the currently editable Behavior Tree nodes and non-editable Subtree nodes.
Added a lock/unlock toggle next to the "Add Keepout Zone" button to disable unintentional user interaction with collision keepout zones.
tip_links in PlanCartesianPath and other Behaviors requiring tip_links as input.The Joint Trajectory Admittance Controller (JTAC), and corresponding Behavior ExecuteTrajectoryWithAdmittance, can now be used without defining the admittance parameters.
In this case, the controller will execute the trajectory without admittance control, making it equivalent to a standard Joint Trajectory Controller.
This makes it easier to use the JTAC in applications that do not require admittance control.
The Path IK MTC Stage (e.g. SetupMTCPathIK) now generates partial trajectories, for visualization purposes, when the full path is not feasible. This allows users to see how far along the path the IK solver was able to compute valid waypoints before encountering an issue such as joint limits or collisions.
Fixed Tutorial 1 to use the updated name ML Segment Image Objective and image.
joint_names field in the JointTrajectory message and reorders joint values accordingly.RecordJointTrajectory improved to better handle multithreaded service callbacks and creation to prevent race conditions.