Skip to main content

5.1.6 Version Release Notes

tros-humble​

Version: 2.4.5 (2025-10-28)​

Bug Fixes (RDK X5 Platform):

  • Fixed the bug of image resize acceleration with vse in the hobot_cv image processing acceleration module.
  • Fixed the incorrect inference latency statistics in the on-board model inference framework dnn_node.
  • Fixed the runtime loading failure caused by incorrect installation paths of ros component so in modules such as MIPI image capture and algorithm packages.
  • Optimized the configuration of Image Codec, removed invalid configuration parameters, and added debug-specific configuration parameters.

Version: 2.4.4 (2025-10-24)​

New Features (RDK S100 Platform):

  • Added ROI inference mode to the on-board algorithm inference framework dnn_node; the number of output output tensor is output_size x roi_size.
  • Added the algorithm example of Human Detection and Tracking (Ultralytics YOLO Pose).
  • Added the algorithm example of Human Instance Tracking: extracts human features based on the reid model, and stores, manages and queries features via the SQlite database. Disabled the ROI-based human tracking mode of the human detection and tracking node mono2d_body_detection in the launch file; restricted the input ROI size to be less than 3.5 times the actual model input size.
  • Added the algorithm example of Hand Keypoint and Gesture Recognition (mediapipe), which implements basic palm detection, pre/post-processing, and ai msg publishing. Supports MIPI cameras, USB cameras, and local image playback; supports zero-copy and non-zero-copy image acquisition.
  • Added the algorithm example of DeepSeek Large Language Model, supporting human-machine dialogue. RDK S100 and RDK S100P added support for the DeepSeek_R1_Distill_Qwen_1.5B and DeepSeek_R1_Distill_Qwen_7B models.

Version: 2.4.3 (2025-09-15)​

New Features (RDK X5 Platform):

Version: 2.4.2 (2025-08-29)​

New Features (RDK S100 Platform):

  • MIPI Image Capture supports the 230ai stereo module.
  • Object Detection YOLO supports the yolo11 and yolov12 algorithms; added scripts for startup in component mode, supporting 4K image capture for inference via mipi cam.
  • EdgeSAM Segment Anything added edgesam for model inference, nv12 format data input, and padding for segmentation results (for co-visualization in stereo depth cases).
  • Added the Text-Image Feature Retrieval Algorithm for text-image feature extraction and retrieval.
  • Added the DOSOD Algorithm: added the self-developed open-vocabulary detection DOSOD edge deployment package, and a re-parameterization quantization method for custom modification of detection categories in the DOSOD model.
  • Added Stereo OCC, integrating the self-developed stereo OCC network.
  • Vision-Language Model added adaptation for the smolvlm2 model (supporting image playback and subscription mode), and the ability to output complete topics after LLM model inference.
  • Added the Stereo Depth Estimation Algorithm.

Version: 2.4.1 (2025-07-30)​

New Features (RDK X5 Platform):

  • MIPI Image Capture supports the imx415 module.
  • EdgeSAM Segment Anything added edgesam for model inference, nv12 format data input, and padding for segmentation results (for co-visualization in stereo depth cases).
  • Added the Human Instance Tracking Algorithm: extracts human features based on the reid model, and stores, manages and queries features via the SQlite database.
  • Added the Stereo OCC Algorithm, integrating the self-developed stereo OCC network.
  • Vision-Language Model added adaptation for the smolvlm2 model (supporting image playback and subscription mode), and the ability to output complete topics after LLM model inference.

Version: 2.4.0 (2025-05-12)​

New Features:

  • Supported the RDK S100 platform.

Version: 2.3.3 (2025-04-30)​

New Features:

  • Supported the RDK X5 Module platform.
  • Added an open-source ASR Solution based on sensevoice_cpp, supporting push of command words and ASR data.
  • Optimized the post-processing latency of the Stereo Depth Estimation Algorithm and added the V2.3 model version.
  • Added an edge-side Vision-Language Model algorithm example based on llama.cpp.

Version: 2.3.2 (2025-01-15)​

Feature Changes:

New Features:

  • Added ZED Camera Image Capture to start the ZED camera for stereo image acquisition (as input for the stereo depth estimation algorithm).
  • Added the DOSOD Algorithm and the self-developed open-vocabulary detection DOSOD edge deployment package.

Bug Fixes:

Version: 2.3.1 (2024-11-20)​

Feature Changes:

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

New Features:

Bug Fixes:

Version: 2.3.0 (2024-09-19)​

New Features:

Version: 2.2.0 (2024-04-11)​

Feature Changes:

  • Adapted to Ubuntu 22.04 and ROS2 Humble based on TROS Foxy 2.1.3.
  • Changed the TROS installation path from /opt/tros to /opt/tros/humble, aligning with the directory hierarchy and naming of ROS2 installation paths.
  • Discontinued the tros-ros-base installation package (including ROS2 core packages such as rclcpp, rclpy, ros2cli); uses standard ROS2 distribution packages, and dependent ROS2 Humble packages are installed automatically when installing TROS Humble.
  • Adopted ROS2 FastDDS zero-copy communication for modules involving image data (data acquisition, image codec, algorithm examples, etc.).
  • Changed the QoS Reliability for zero-copy communication from RMW_QOS_POLICY_RELIABILITY_RELIABLE (rclcpp::QoS()) to RMW_QOS_POLICY_RELIABILITY_BEST_EFFORT (rclcpp::SensorDataQoS()) to avoid potential stability risks in zero-copy mode.
  • Refactored hobot_dnn to use the underlying on-board inference framework libdnn instead of easydnn.
  • Upgraded the speech algorithm SDK in hobot_audio to use the underlying on-board inference framework libdnn instead of easydnn.
  • Adapted hobot_trigger to ROS2 Humble rosbag2.

New Features:

  • Added bloom compilation and packaging scripts to robot_dev_config for compiling and packaging TROS on the ARM platform.
  • Added the frame_ts_type configuration item to the hobot_mipi_cam node, supporting realtime (for calculating communication latency) and sensor (default, for sensor timestamp synchronization) parameters.
  • Added the hobot_shm node for configuring the ROS2 zero-copy environment.

Bug Fixes:

  • Fixed compatibility issues introduced by compiler upgrades.
  • Fixed path dependency issues in compiling some ROS2 packages on the board.

tros-foxy​

Version: 2.1.3 (2024-03-11)​

Feature Changes:

  • Changed the data type of JPEG-compressed images from sensor_msgs::msg::Image to the standard sensor_msgs::msg::CompressedImage, supporting viewing TROS-published JPEG images with tools such as Foxglove and ROS2 RQT (involving hobot_websocket, hobot_codec, hobot_image_publisher, hobot_usb_cam modules).
  • Unified the use of the jpeg/mjpeg configuration item to specify publishing/subscribing JPEG-compressed images; removed the jpeg-compressed/mjpeg-compressed configuration items (involving hobot_codec and hobot_usb_cam modules).
  • Introduced the TROS_DISTRO environment variable for indicating the TROS distribution; the value of TROS_DISTRO is empty after executing source /opt/tros/setup.bash/source /opt/tros/local_setup.bash. Changed the configuration file path for modules such as hobot_codec, hobot_audio, hobot_mipi_cam, and hobot_usb_cam from /opt/tros/lib to /opt/tros/${TROS_DISTRO}/lib.

Version: 2.1.2 (2024-01-19)​

New Features:

  • Refactored hobot_usb_cam to support more format configurations and transcoding.
  • Updated the speech SDK in hobot_audio to support both 2-mic and 4-mic microphone boards; added the micphone_name configuration for device ID.

Bug Fixes:

  • Fixed the incorrect step field setting in data messages sent by the hobot_rgbd_cam node.
  • Updated the audio playback function call in hobot_tts to resolve playback failures on new system versions.
  • Removed the config device tree file from hobot_llm and updated the README; the ION memory size can be set via command-line tools on new system versions.

Version: 2.1.1 (2023-11-03)​

New Features:

  • Added the hobot_chatbot node: calls the intelligent speech, large language model, and text-to-speech modules to implement on-board voice chat functionality.

Bug Fixes:

  • Fixed the application exit caused by specific characters in the text-to-speech hobot_tts node.

Version: 2.1.0 (2023-09-14)​

Feature Changes:

  • Updated tros-ros-base to the latest ROS2 Foxy source code, compatible with the latest ROS2 Foxy packages.
  • Only source /opt/tros/setup.bash is required to use ROS2 Foxy packages; no more script-based soft link creation.

New Features:

  • Added a parameter to the text-to-speech hobot_tts node for specifying the audio playback device.
  • Added the large language model hobot_llm node for edge-side LLM experience.
  • Added the jpeg-compressed configuration item to the in_format parameter of the Image Codec hobot_codec node, and the subscribed topic data type is selected based on the configuration item.

Bug Fixes:

  • Fixed the incorrect step field setting in RGB format data messages sent by the MIPI image capture hobot_mipi_cam node.

Version: 2.0.2 (2023-08-28)​

Feature Changes:

  • Changed the ROS2 source configured during tros.b installation (/etc/apt/sources.list.d/ros2.list) to the Tsinghua mirror source, resolving slow installation and failure of ROS2 packages.

New Features:

  • Added permission check when configuring the environment via tros.b startup scripts (source /opt/tros/setup.bash and source /opt/tros/local_setup.bash). Automatically switches to the root account if the current account does not have root privileges, resolving tros.b usage failures due to insufficient permissions.
  • Added audio device ID parameter configuration to the intelligent speech algorithm hobot_audio node for easy secondary development.
  • Added the ability to send tasks to the Trigger module via std_msg topics in the event trigger hobot_trigger node, standardizing the Trigger configuration method.

Bug Fixes:

  • Fixed the processing failure of simultaneous crop&resize in the image acceleration processing hobot_cv node.
  • Fixed the error log output during startup of the MIPI image capture hobot_mipi_cam node.
  • Fixed the invalid configuration of the launch file for the data visualization message conversion hobot_visualization node.

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

Feature Changes:

  • Upgraded the speech algorithm to optimize ASR (Automatic Speech Recognition) performance.
  • Optimized the model_name configuration item of algorithm examples: automatically parses the model_name configuration from the model file, resolving model loading failures caused by incorrect parameter configuration and improving the ease of secondary development for algorithms.
  • Removed the nav2 package from the tros.b installation package; users can install the latest ROS2 nav2 package directly on the RDK via the apt command, resolving stability issues with the old nav2 version.

New Features:

  • Added support for the RDK Ultra platform.
  • Added nodes such as hobot_trigger and hobot_visalization for Trigger event triggering and rosbag data acquisition/visualization, helping users locate, reproduce and visualize perception, planning and control issues in robot scenarios. Users can also perform secondary development to implement data triggering, recording and real-time transmission.
  • The USB image capture node automatically adapts to the device ID of USB cameras, lowering the threshold for using USB cameras.
  • Added the Visual Inertial Odometry (VIO) algorithm node: implements a low-cost, high-robustness high-precision robot positioning algorithm based on vision.
  • Added the text-to-speech hobot_tts node for converting text to speech for broadcast.
  • Added the LiDAR object detection algorithm hobot_centerpoint node.
  • Added the BEV perception algorithm hobot_bev node.
  • Added the stereo depth estimation algorithm hobot_stereonet node.

Bug Fixes:

  • Upgraded easydnn (v1.6.1) and dnn (v1.18.4) for RDK X3, fixing operator crash issues and adding support for more operators.
  • Fixed incorrect depth data published by the RGBD image capture node.

Other Updates:

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

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 the 1.x version of tros.b are recommended to upgrade to the 2.x version.

Feature Changes:

  • Changed the code hosting platform from GitLab to GitHub for easier secondary development by more developers.
  • Integrated a more efficient package management mechanism to speed up version upgrades and simplify robot application installation.

New Features:

  • Supported the new core board development kit RDK X3 Module.
  • Added speech ASR recognition result output to hobot_audio for easy development of speech applications.

Bug Fixes:

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

Other Updates:

  • Updated the launch scripts of examples: applications reference the launch scripts of dependent modules and configure parameters.
  • Updated the D-Robotics logo on the display side of websocket.
Last updated on