ROS 1 Bridge Workshop#

This tutorial connects a ROS 1 Noetic application to Robotics AI Suite inference on ROS 2 Jazzy, through ros1_bridge. The suite supports only ROS 2, and Noetic and ros1_bridge packages exist only for Ubuntu 20.04, so the tutorial uses two workshops:

Workshop

Base

Runs

ros1-bridge

Ubuntu 20.04

ROS 1 Noetic (roscore and a camera node) and the ROS 2 Foxy ros1_bridge, from packages.ros.org

ros1-detector

Ubuntu 24.04

The Robotics AI Suite on Jazzy, and the object detection tutorial’s SSD model with OpenVINO™ on the CPU, GPU or NPU

        flowchart LR
  subgraph R1["ros1-bridge (Ubuntu 20.04)"]
    cam["ros1_camera.py<br/>(rospy)"] -- "/ros1/image" --> b["parameter_bridge<br/>(Foxy)"]
    b -- "/ros1/detections" --> cam
  end
  subgraph R2["ros1-detector (Ubuntu 24.04)"]
    det["ros1_detector.py<br/>(rclpy + OpenVINO)"]
  end
  b <-- "DDS, domain 42<br/>workshopbr0" --> det
  det -. GPU / NPU .- hw[(Host GPU / NPU)]
    

The ROS 1 camera node publishes a raw RGB frame (sensor_msgs/Image) at a fixed rate. The bridge carries it to ROS 2, where the detector runs SSD MobileNet v2 and publishes the detections as JSON (std_msgs/String). The bridge carries them back, and the camera node measures the round trip for each frame.

Prerequisites#

Launch the workshops#

  1. Launch both workshops. ros1-bridge takes about 2 minutes; ros1-detector installs the full suite and takes about 11 minutes:

    cd src/workshops/ros1-bridge
    workshop launch ros1-bridge
    workshop launch ros1-detector
    

    Workshop warns that Ubuntu 20.04 bases are no longer supported. Acknowledge the warning with workshop okay; see Known issues.

  2. To run detection on the NPU, connect it to the detector:

    workshop connect ros1-detector/robotics-ai-suite:npu :custom-device
    

Run the round trip#

The check script runs both sides from the host, and prints PASS if at least 95% of the frames get their detections back:

cd ..
./scripts/ros1-bridge-check.sh                      # GPU, 640 x 480 at 10 Hz, for 30 s
./scripts/ros1-bridge-check.sh NPU 1280 720 30 30   # NPU, 1280 x 720 at 30 Hz, for 30 s

Its arguments are:

scripts/ros1-bridge-check.sh [device] [width] [height] [hz] [seconds] [rmw]

Argument

Default

Description

device

GPU

CPU, GPU or NPU

width, height

640, 480

Frame size. The frame is generated on first use

hz

10

Publish rate

seconds

30

Measured time, after an 8-second warm-up

rmw

rmw_cyclonedds_cpp

ROS 2 middleware on both sides. rmw_fastrtps_cpp also works

Example results#

Note

Individual results may vary.

On an Intel® Core™ Ultra X7 358H processor with Cyclone DDS, every run answered 100% of the frames, with the same labels (bicycle, car):

Input

Device

Round trip, median

Round trip, 90th percentile

Resize and inference, median

640 x 480 at 10 Hz

GPU

4.7 ms

5.3 ms

1.68 ms

1280 x 720 at 30 Hz

NPU

10.8 ms

12.2 ms

3.83 ms

1280 x 720 at 30 Hz

GPU, Fast DDS

11.5 ms

15.7 ms

1.77 ms

Latencies vary between runs; repeated GPU runs at 640 x 480 and 10 Hz gave a median round trip of 4.7 to 8.4 ms. Compare devices from runs made back to back. Compared with running the camera on ROS 2 in the same workshop as the detector, the bridge adds about 2 to 4 ms per round trip.

Run each side by hand#

The check script uses these actions. To run them yourself, use two terminals, and start the detector first:

Workshop

Action

Description

ros1-detector

frames [width] [height]

Writes the tutorial’s coco_bike.jpg as a raw RGB frame to ros1-bridge/frames/

ros1-detector

detect [device] [max_seconds] [rmw]

The detector on ROS 2 domain 42. Prints ready once the model is compiled, and exits 5 seconds after the last frame

ros1-bridge

bridge-detect [width] [height] [hz] [seconds] [rmw]

roscore, parameter_bridge for /ros1/image and /ros1/detections, then the ROS 1 camera node. Returns 1 if fewer than 95% of frames are answered

Design notes#

  • parameter_bridge, not dynamic_bridge. dynamic_bridge crashes when it reads the graph information of Jazzy nodes. parameter_bridge bridges a fixed list of topics, set with rosparam set /topics, and works.

  • ROS 2 Foxy for the bridge. ros-foxy-ros1-bridge is the only packaged ros1_bridge; there is no package for Humble or Jazzy. Both Cyclone DDS and Fast DDS carry sensor_msgs/Image and std_msgs/String between Foxy and Jazzy. Types whose definitions changed between Foxy and Jazzy may not work; see Humble and Jazzy on the same domain for the same effect between Humble and Jazzy.

  • ROS domain 42. Both sides use ROS_DOMAIN_ID=42, so the loop doesn’t mix with other workshops on the default domain 0.

  • Cyclone DDS configuration. The detector unsets CYCLONEDDS_URI, because the suite sets it to a file that doesn’t exist; see Troubleshooting.

  • Resize on the CPU. The detector resizes frames with OpenCV. With OpenVINO™’s embedded resize, the NPU returns no detections; this is the same problem as in the Object Detection workshop.

  • Consumers on the host. A ROS 2 node on the host that uses Cyclone DDS needs the ros1-bridge workshop’s IP address as a unicast peer to see the bridged topics.

Known issues#

  • Ubuntu 20.04 base. Workshop launches it, but warns that 20.04 bases are no longer supported and will need 22.04 or later before its next release. Noetic and ros1_bridge packages exist only for 20.04, and Noetic and Foxy are both end of life. If Workshop drops 20.04 bases, the alternative is ROS-O, Noetic-based ROS 1 packages for Ubuntu 22.04 and 24.04, with a bridge built from source.

Source#

The definitions, nodes and package list are in src/workshops/ros1-bridge/. The check script is src/workshops/scripts/ros1-bridge-check.sh.