Big news – we've got a new release of DepthAI ROS packages out, built for ROS 2 Kilted and using DepthAI V3.
Core Updates
- DepthAI Library V3 Integration
The underlying DepthAI library has been updated to V3, bringing numerous new features and bug fixes. For detailed information, refer to the depthai-core repository.
This is the first release that supports OAK4 devices (OAK4-D and OAK4-S)
- Renamed Nodes and Parameters
The main node in depthai_ros_driver is now driver (formerly camera). Users should update their YAML configurations to reflect this change.
camera.launch.py has been deprecated in favor of driver.launch.py.
i_mx_id has been replaced by i_device_id.
The process of creating Neural Networks has been streamlined in both depthai_examples and depthai_ros_driver.
i_nn_family now corresponds to i_nn_kind (note: this parameter may be renamed in the future).
i_nn_model is used to pass the actual NN model.
New Features and Capabilities:
- Improved RGBD and Pointcloud Functionality
A dedicated RGBD Node and Pointcloud converter have been added for generating colored pointclouds with significantly improved performance.
An RGBD node is now available for publishing RGBD pointclouds, eliminating the need for external nodes and boosting performance.
Thermal Node
A new Thermal node has been introduced for devices equipped with thermal sensors.
Camera Nodes for Enhanced Stream Control
Mono/Color camera nodes have been replaced by more versatile Camera nodes.
This allows for requesting undistorted streams and specifying exact width and height without ISP scaling concerns.
Undistorted streams can be requested by setting i_undistorted and choosing i_resize_mode.
Default camera output size is now 640x400 for better performance.
IMU Publishing in RDF Frame: IMU data is now consistently published in the RDF frame.
TFPublisher by Default
The system now defaults to using TFPublisher instead of URDF descriptions for more accurate results.
Host nodes can now be integrated into the pipeline for improved performance.
The Stereo node now features an independent fps parameter, separate from the sensor FPS.
Experimental - New VIO node - based on open-source libraries (Basal), running currently on host with RVC2 Devices (RVC4 support coming soon!), publishing position and odometry information
We also plan to release a VSLAM node that is based off an open source RTABMap library soon! (It is currently available on base depthai-core library in both C++ and Python implementations)
Refinements and Optimizations
- Updated Socket/Frame Naming
Socket and frame naming conventions have been updated to accurately reflect different devices and configurations. In rs_mode, names align with Realsense, while specific devices (e.g., OAK-T, OAK-D-SR-PoE) use sensor-specific names (e.g., tof instead of rgb).
- Simplified depthai_examples
The depthai_examples repository has been extensively modified to remove deprecated examples and simplify existing code. Current examples include:
rgbd_spatial_detections
feature_tracker
converter_to_cv
imu_publisher
rgb_publisher
rgbd_publisher
rgb_subscriber
rgb_compressed_publisher
disparity_publisher
detection_publisher
tof_publisher
tof_rgbd
thermal_publisher
Parameter Changes for Pipeline Creation
Parameters related to pipeline creation have been moved to the PipelineGen PH, including:
i_pipeline_type
i_nn_type
i_enable_rgbd
Rviz and TF Frame Updates
New Rviz configurations have been added for several examples.
TF Frames for base_frame and parent_frame have been adjusted to NodeName and oak_parent_frame, respectively.
For a full changelog you can refer to this link.
How to get it:
Follow instructions at our official documentation page.
Heads up on some known issues
- Segmentation networks, VIO and SLAM only work on RVC2 devices for now.
- To use NNs you need internet connection (for model downloading)
- For RVC4 devices, you can only use COPY as ImuSyncMethod.
- Setting image orientation isn't supported yet.
- Not all examples in depthai_filters work properly yet.
- RGBD node doesn't have GPU support yet.
- Using external topics as inputs to the driver isn’t supported yet
