Skip to main content

5.1.6 Release Notes

tros-jazzy​

Version: 2.5.4 (2026-05-19)​

Changes (RDK S600 platform):

  • Adapted to system SDK RKD S600 v5.1.0.

Version: 2.5.3 (2026-05-08)​

New Features (RDK S100 Platform):

  • Added mono_edgetam_prompt: Supports two prompt styles via prompt_mode: point prompts or box prompts. Exports prompt initialization results (memory feature file) for downstream tracking, and provides optional local rendering output.
  • Added mono_edgetam_track to load features and use them as tracking initialization. Supports continuous tracking in both local image and subscription stream modes. Updates tracking memory frame by frame and publishes segmentation/tracking results.

Feature Changes (RDK S100 Platform):

  • Adapted to system version: RDKS100_V4.0.6.

Version: 2.5.1 (2026-03-10)​

Changes:

  • Adapted to system SDK RKD S600 v5.0.1.
  • Refactored mipi_cam videobuff management from stream capture to message publishing, as well as the stitching thread; changed calibration info topics for stitched images to "image_combine_raw/left/camera_info" and "image_combine_raw/right/camera_info".

New Features:

  • Added detection algorithms yolov2, yolov3, yolov5; segmentation algorithm unet; classification algorithm mobilenet.
  • Added open-vocabulary object detection algorithm DOSOD.
  • Added human detection and tracking algorithm (Ultralytics YOLO Pose).
  • Added human instance tracking algorithm reid.
  • Added segment-anything algorithm MobileSAM.

Version: 2.5.0 (2025-12-02)​

New Features:

  • Support for RDK S600 platform.

tros-humble​

Version: 2.5.3 (2026-05-08)​

New Features (RDK S100 platform):

  • Added mono_edgetam_prompt: supports two prompt styles via prompt_mode—point prompts or box prompts. Exports prompt initialization results (in-memory feature files) for downstream tracking, with optional local rendering output.
  • Added mono_edgetam_track to load features and use them for tracking initialization. Supports continuous tracking in local image and subscribed stream modes. Updates tracking memory frame by frame and publishes segmentation/tracking results.

Changes (RDK S100 platform):

  • Adapted to system version: RDKS100_V4.0.6.

Version: 2.5.2 (2026-03-17)​

New Features (RDK X5 platform):

  • Added epipolar alignment detection to the stereo depth estimation algorithm; adjusted subscribed messages per hobot_mipi_cam to "image_combine_raw/left/camera_info" and "image_combine_raw/right/camera_info"
  • Stereo OCC algorithm supports mipi cameras.
  • Object detection algorithm supports yolo26.
  • sensevoice_ros2 algorithm adds configuration options for Chinese/English mode.
  • Added hand keypoint and gesture recognition algorithm based on palm_detection_mediapipe and hand_landmarks_mediapipe.

Changes (RDK X5 platform):

  • Adapted to system version: RDK 3.5.0 (Linux SDK V1.1.2)
  • Refactored mipi_cam videobuff management from stream capture to message publishing and the stitching thread. Changed calibration info topics for stitched images to "image_combine_raw/left/camera_info" and "image_combine_raw/right/camera_info". Refactored X5 eeprom read handling. Added subdirectory publishing on X5. Added combined 132gs imu data publishing.

Version: 2.4.6 (2026-02-02)​

Changes (RDK S100 platform):

  • Adapted to system version V4.0.5 and OE version V3.7.0.

Version: 2.4.5 (2025-10-28)​

Bug Fixes (RDK X5 platform):

  • Fixed a bug in the image processing acceleration module hobot_cv when using vse to accelerate image resize.
  • Fixed incorrect inference latency statistics in the on-board model inference framework dnn_node.
  • Fixed runtime loading failures caused by incorrect installation paths for ros component so in MIPI image capture and algorithm modules.
  • Optimized image codec configuration: removed invalid configuration parameters and added debug configuration parameters.

Version: 2.4.4 (2025-10-24)​

New Features (RDK S100 platform):

  • On-board algorithm inference framework dnn_node adds ROI inference mode; the number of output output tensors is output_size x roi_size.
  • Added human detection and tracking (Ultralytics YOLO Pose) algorithm example.
  • Added human instance tracking algorithm example, which extracts human features using the reid model and stores, manages, and queries features via an SQlite database. The launch file disables ROI-based human tracking in the human detection and tracking node mono2d_body_detection; input ROI size is limited to less than 3.5x the actual model input size.
  • Added hand keypoint and gesture recognition (mediapipe) algorithm example. Implements basic palm detection, pre/post-processing, and ai msg publishing; supports mipi camera, usb camera, and local image replay; supports zero-copy and non-zero-copy image acquisition.
  • Added DeepSeek large language model algorithm example. Supports human-machine dialogue; RDK S100 and RDK S100P newly support DeepSeek_R1_Distill_Qwen_1.5B and DeepSeek_R1_Distill_Qwen_7B models.

Version: 2.4.3 (2025-9-15)​

New Features (RDK X5 platform):

Version: 2.4.2 (2025-8-29)​

New Features (RDK S100 platform):

  • MIPI image capture supports 230ai stereo module.
  • YOLO object detection supports yolo11 and yolov12 algorithms; added component-based launch scripts supporting 4K image inference with mipi cam.
  • EdgeSAM segment anything algorithm adds edgesam for model inference; adds nv12 format data input; adds padding for segmentation results for joint visualization in stereo depth use cases.
  • Added text-image feature retrieval algorithm for text-image feature extraction and retrieval.
  • Added DOSOD algorithm, D-Robotics open-vocabulary detection DOSOD on-device deployment package; added re-parameterization quantization method for customizing detection categories in DOSOD models.
  • Added stereo OCC, integrating D-Robotics stereo OCC network.
  • Vision-language model algorithm adds smolvlm2 model support, with image replay and subscription modes; adds ability to output complete topics after llm model inference completes.
  • Added stereo depth estimation algorithm.

Version: 2.4.1 (2025-7-30)​

New Features (RDK X5 platform):

  • MIPI image capture supports imx415 module.
  • EdgeSAM segment anything algorithm adds edgesam for model inference; adds nv12 format data input; adds padding for segmentation results for joint visualization in stereo depth use cases.
  • Added human instance tracking algorithm, which extracts human features using the reid model and stores, manages, and queries features via an SQlite database.
  • Added stereo OCC algorithm, integrating D-Robotics stereo OCC network.
  • Vision-language model adds smolvlm2 model support, with image replay and subscription modes; adds ability to output complete topics after llm model inference completes.

Version: 2.4.0 (2025-5-12)​

New Features:

  • Support for RDK S100 platform.

Version: 2.3.3 (2025-4-30)​

New Features:

Version: 2.3.2 (2025-1-15)​

Changes:

New Features:

  • Added ZED camera image capture to launch ZED camera for stereo images as input to stereo depth estimation algorithm.
  • Added DOSOD algorithm, D-Robotics open-vocabulary detection DOSOD on-device deployment package.

Bug Fixes:

Version: 2.3.1 (2024-11-20)​

Changes:

  • Upgraded dependent opencv version from 3.4.5 to 4.5.4 (latest release version used on Ubuntu 22.04).

New Features:

Bug Fixes:

Version: 2.3.0 (2024-09-19)​

New Features:

Version: 2.2.0 (2024-04-11)​

Changes:

  • Based on TROS Foxy 2.1.3, adapted to Ubuntu 22.04 and ROS2 Humble.
  • TROS installation path changed from /opt/tros to /opt/tros/humble, consistent with ROS2 installation path hierarchy and naming.
  • No longer provides tros-ros-base package (includes rclcpp, rclpy, ros2cli and other ROS2 base packages); uses standard ROS2 distribution packages; installing TROS Humble automatically installs dependent ROS2 Humble.
  • Uses ROS2 fastdds zero-copy communication for modules involving image data such as data acquisition, image codec, and algorithm examples.
  • Zero-copy communication QoS Reliability changed from RMW_QOS_POLICY_RELIABILITY_RELIABLE (rclcpp::QoS()) to RMW_QOS_POLICY_RELIABILITY_BEST_EFFORT (rclcpp::SensorDataQoS()) to avoid potential stability risks when using zero-copy.
  • Refactored hobot_dnn to use lower-level on-board inference framework libdnn instead of easydnn.
  • hobot_audio upgraded speech algorithm SDK to use lower-level on-board inference framework libdnn instead of easydnn.
  • hobot_trigger adapted to ROS2 Humble version rosbag2.

New Features:

  • robot_dev_config added bloom build and packaging scripts for ARM platform TROS compilation and packaging.
  • hobot_mipi_cam node added frame_ts_type configuration option, supporting realtime (for communication latency calculation) and sensor (default, for sensor timestamp synchronization) parameters.
  • Added hobot_shm node for configuring ROS2 zero-copy environment.

Bug Fixes:

  • Fixed compatibility issues introduced by compiler upgrade.
  • Fixed path dependency issues in some ROS2 pkg on-board compilation.

tros-foxy​

Version: 2.1.3 (2024-03-11)​

Changes:

  • JPEG compressed image data type changed from sensor_msgs::msg::Image to standard sensor_msgs::msg::CompressedImage, enabling tools such as foxglove and ros2 rqt to view TROS-published JPEG images. Affects hobot_websocket, hobot_codec, hobot_image_publisher, hobot_usb_cam modules.
  • Unified use of jpeg/mjpeg configuration options for publishing/subscribing JPEG compressed images; removed jpeg-compressed/mjpeg-compressed configuration options. Affects hobot_codec and hobot_usb_cam modules.
  • Introduced TROS_DISTRO environment variable to indicate TROS distribution. After running source /opt/tros/setup.bash/source /opt/tros/local_setup.bash, the TROS_DISTRO environment variable is empty. Configuration file paths for hobot_codec, hobot_audio, hobot_mipi_cam, hobot_usb_cam and other modules changed from /opt/tros/lib to /opt/tros/${TROS_DISTRO}/lib.

Version: 2.1.2 (2024-01-19)​

New Features:

  • Refactored hobot_usb_cam with support for more format configurations and transcoding.
  • hobot_audio updated speech SDK, supporting both 2mic and 4mic microphone boards; added micphone_name configuration for device ID.

Bug Fixes:

  • hobot_rgbd_cam node fixed incorrect step field in sent data messages.
  • hobot_tts updated audio playback function calls to fix playback failures on newer system versions.
  • hobot_llm removed config device tree file and updated README; newer system versions can set ION memory size via command-line tools.

Version: 2.1.1 (2023-11-03)​

New Features:

  • Added hobot_chatbot node that invokes intelligent speech, large language model, and text-to-speech modules to enable on-board voice chat.

Bug Fixes:

  • Text-to-speech hobot_tts node fixed application exit caused by certain characters.

Version: 2.1.0 (2023-09-14)​

Changes:

  • tros-ros-base updated to latest ROS2 foxy source, compatible with latest ROS2 foxy packages.
  • Using ROS2 foxy packages only requires source /opt/tros/setup.bash; no longer need scripts to create symlinks.

New Features:

  • Text-to-speech hobot_tts node added parameter to specify audio playback device.
  • Added large language model hobot_llm node for on-device LLM experience.
  • Image codec hobot_codec node configuration parameter in_format added jpeg-compressed option; subscribes to topic data type based on configuration option.

Bug Fixes:

  • MIPI image capture hobot_mipi_cam node fixed incorrect step field in RGB format data messages.

Version: 2.0.2 (2023-08-28)​

Changes:

  • ROS2 source configured during tros.b installation (/etc/apt/sources.list.d/ros2.list) changed to Tsinghua mirror, fixing slow and failed ROS2 package installation.

New Features:

  • Added permission check when tros.b startup scripts configure the environment (source /opt/tros/setup.bash and source /opt/tros/local_setup.bash). If the current account lacks root permissions, automatically enters the process to switch to root account, fixing tros.b usage failures due to insufficient permissions.
  • Intelligent speech algorithm hobot_audio node added audio device number parameter configuration for easier secondary development.
  • Event trigger hobot_trigger node added ability to assign tasks to Trigger module via std_msg topics, standardizing Trigger configuration.

Bug Fixes:

  • Fixed image acceleration processing hobot_cv node failure when performing crop&resize simultaneously.
  • Fixed MIPI image capture hobot_mipi_cam node error log output at startup.
  • Fixed invalid launch file configuration for data visualization message conversion hobot_visualization node.

Version: 2.0-Release (2.0.1) (2023-06-10)​

Changes:

  • Upgraded speech algorithm to optimize ASR (Automatic Speech Recognition) performance.
  • Optimized algorithm example model_name configuration option to automatically parse model_name from model files, fixing model loading failures caused by incorrect parameter configuration and improving ease of secondary development.
  • tros.b installation package no longer includes nav2 packages; users install the latest ROS2 nav2 packages directly on RDK via apt, fixing stability issues in older nav2 versions.

New Features:

  • Added hobot_trigger and hobot_visalization nodes for Trigger event triggering, rosbag data acquisition and visualization, helping users locate, reproduce, and visualize perception, planning, and control issues in robot scenarios. Users can also develop data triggering, recording, and real-time transmission capabilities.
  • USB image capture node auto-adapts USB camera device numbers, lowering the barrier for USB camera usage.
  • Added Visual Inertial Odometry (VIO) algorithm node for low-cost, robust, high-precision robot localization based on vision.
  • Added text-to-speech hobot_tts node to convert text to speech for playback.
  • Added LiDAR object detection algorithm hobot_centerpoint node.
  • Added BEV perception algorithm hobot_bev node.
  • Added stereo depth estimation algorithm hobot_stereonet node.

Bug Fixes:

  • Upgraded RDK X3 easydnn (version 1.6.1) and dnn (version 1.18.4), fixing operator crash issues and supporting more operators.
  • Fixed incorrect depth data published by RGBD image capture node.

Other Updates:

  • Optimized human detection and tracking algorithm node to adapt output perception result coordinates based on input image resolution.
  • Fixed orb_slam3 algorithm compilation failure caused by incorrect compilation script path.

Version: 2.0-Beta (2.0.0) (2023-05-29)​

2.0-Beta (2.0.0) is the first 2.x version of tros.b. Users of tros.b 1.x are recommended to upgrade to 2.x.

Changes:

  • Code hosting platform moved from Gitlab to GitHub for easier secondary development by more developers.
  • Integrated more efficient package management to speed up version upgrades and simplify robot application installation.

New Features:

  • Support for the new RDK X3 Module core board development kit.
  • hobot_audio added ASR recognition result output for voice application development.

Bug Fixes:

  • Fixed dnn_node built-in MobileNet_SSD model post-processing crash in multi-threaded scenarios.
  • Fixed dnn_node DDR input model inference failure on X86 platform
  • Fixed hobot_codec and hobot_image_publisher compilation failures on X86 platform.

Other Updates:

  • Updated example launch scripts to reference dependent module launch scripts and configure parameters.
  • websocket updated display-side D-Robotics logo.
Last updated on