RealSense2 Tutorial Camera Benchmark#

This benchmark measures KPI performance of the RealSense Camera with ROS 2 Sample Application — a live-camera streaming pipeline (camera driver + rviz2), supporting both a USB-connected RealSense camera and a GMSL-connected RealSense Depth Camera D457.

Unlike the simulation benchmarks, this is a live-camera scenario: there’s no goal-based stop condition, so the Robotics System Profiler runs for a fixed duration, capturing timing, resource, and optionally GPU metrics.

Note: Currently USB cameras are the only supported cameras.

Prerequisites#

Complete the Installation Guide and ensure the RealSense tutorial runs successfully before benchmarking. Run scripts/find_cameras.sh (from the realsense2_tutorial package) first to confirm whether a USB or GMSL camera is connected.

Single Run#

A single run launches realsense2_tutorial.launch.py, waits for the camera driver and rviz2 to initialize, then monitors for a fixed timeout before stopping cleanly.

# USB camera (default)
bash src/realsense2_tutorial_run.sh

# USB camera + record a KPI rosbag
bash src/realsense2_tutorial_run.sh --record

# GMSL D457 camera
bash src/benchmark_runner.sh --run-config config/realsense2_tutorial_gmsl_run.yaml --record

Results land in monitoring_sessions/realsense2_tutorial/<timestamp>/ (or realsense2_tutorial_gmsl/<timestamp>/ for the GMSL run) and can be visualized immediately:

uv run src/visualize_timing.py monitoring_sessions/realsense2_tutorial/<session>/graph_timing.csv --show
uv run src/visualize_graph.py monitoring_sessions/realsense2_tutorial/<session>/graph_timing.csv --show

Parameter

Description

Default

--timeout N

Max duration for the run (seconds)

30

--record

Record KPI topics to a rosbag

—

--plot

Save trigger-timeline PNG plots

—

--output-parent DIR

Session parent directory

monitoring_sessions/realsense2_tutorial/

What the Benchmark Script Does#

realsense2_tutorial_run.sh is a thin wrapper around benchmark_runner.sh using config/realsense2_tutorial_run.yaml. It automates:

  1. Launches ros2 launch realsense2_tutorial realsense2_tutorial.launch.py camera_type:=usb in the background.

  2. Waits init_sleep seconds (20s for USB, 8s for GMSL) before starting the monitor — the RealSense USB camera’s device-detection delay measured ~15-16s even when already connected, so the USB config’s init_sleep accounts for that.

  3. Starts uv run python src/monitor_stack.py to capture graph timing and resource metrics for the configured timeout (default 30s).

  4. Sends SIGINT to stop the camera driver and rviz2 cleanly.

To customize the launch command, camera type, bag topics, or timing, edit config/realsense2_tutorial_run.yaml (USB) or config/realsense2_tutorial_gmsl_run.yaml (GMSL). Do not edit the realsense2_tutorial_run.sh wrapper file.

Visualization#

# Timeline, resource, and frequency plots
uv run src/visualize_timing.py monitoring_sessions/realsense2_tutorial/<session>/graph_timing.csv --show

# Full GPU dashboard (engine busy%, frequency, power)
uv run src/visualize_gpu.py monitoring_sessions/realsense2_tutorial/<session>/gpu_usage.log --show

# Interactive node topology graph
uv run src/visualize_graph.py monitoring_sessions/realsense2_tutorial/<session>/graph_timing.csv --show

Session Data Layout#

monitoring_sessions/
└── realsense2_tutorial/
    └── 20260819_141351/
        ├── session_info.txt
        ├── graph_timing.csv
        ├── resource_usage.json
        ├── gpu_usage.log
        ├── kpi.json
        ├── report.html
        └── visualizations/