ROS 2 MIDDLEWARE
ROS 2, but 5.7x faster.
rmw_hdds is a drop-in replacement for CycloneDDS and FastRTPS. Same API. Zero code changes.
Apex.AI performance_test — Docker — 1 KB @ 1000 Hz — inter-process
18.4 µs
mean latency
5.7x
faster than CycloneDDS
7.5x
faster than FastRTPS
One environment variable. That's it.
No code changes. No recompilation. No new dependencies.
Before — CycloneDDS (default)
# Default ROS 2 ros2 run my_pkg my_node
After — rmw_hdds
export RMW_IMPLEMENTATION=rmw_hdds ros2 run my_pkg my_node
Same topics, same message types, same QoS. Drop-in.
Benchmark Results
Identical conditions. Apex.AI performance_test. Docker. Reproducible.
Mean latency — lower is better
rmw_hdds18.4 µs
5.7x faster
rmw_cyclonedds_cpp104.8 µs
rmw_fastrtps_cpp137.8 µs
| RMW | Mean Latency | Max Latency | vs rmw_hdds |
|---|---|---|---|
| rmw_hdds | 18.4 µs | 62.9 µs | — |
| rmw_cyclonedds_cpp | 104.8 µs | 161.1 µs | 5.7x slower |
| rmw_fastrtps_cpp | 137.8 µs | 165.3 µs | 7.5x slower |
Benchmarks performed with Apex.AI performance_test in Docker containers (inter-process). Results may vary based on hardware and configuration. rmw_hdds supports both intra-process and inter-process modes.
Compatibility
Tested on major ROS 2 distributions.
✓
ROS 2 Humble LTS
Tested
✓
ROS 2 Iron
Tested
✓
ROS 2 Jazzy
Tested
Drop-in replacement for rmw_cyclonedds_cpp and rmw_fastrtps_cpp
Built for Real Robots
Lower latency unlocks tighter control loops and higher sensor frequencies.
👁
Perception Pipelines
Camera detection, object tracking, sensor fusion at high frame rates
🤖
Sensor Fusion
LiDAR + Camera + IMU fusion at 1 kHz without jitter
🧭
Navigation Stack
Nav2 with faster costmap updates and tighter replanning loops
🦾
Robot Arms
Real-time joint control at 1 kHz — sub-millisecond command latency
🚁
Drone Swarms
Multi-agent coordination with low-latency mesh messaging
🏭
Industrial Automation
PLCs, conveyors, AGVs — deterministic timing, zero message loss