Module 17 · Phase 5: Deployment engineering
System Performance & Optimization
Find where latency, CPU, GPU, memory and bandwidth go, and fix message loss with ROS 2 QoS, intra-process communication and executors.
This overview is open. 5 of 5 guides are written; the others show Coming soon until they are complete and tested.
Questions this module answers
- Why is my perception pipeline slow and my CPU high or GPU idle?
- Why are ROS 2 messages dropped or delayed?
- How do I reduce latency and optimize point-cloud processing?
- How do I benchmark the complete system?
What it covers
- End-to-end latency
- CPU, GPU, memory, bandwidth
- QoS, intra-process communication, executors
- Point-cloud and CUDA optimization
- Thermal throttling, storage and logging
Guides and design decisions
Diagnostics, tools and challenges
C. Troubleshooting (5)
- ROS 2 messages are delayedEnd-to-end latency grows or is much higher than the sum of its stages.
- Messages are droppedSubscribers receive fewer messages than the publisher sent.
- CPU usage is too highOne or more cores saturate and the pipeline falls behind.
- GPU utilization is unexpectedly lowThe GPU idles while the CPU is the bottleneck.
- Nodes stop respondingA node stops publishing or serving but the process is alive.
D. Interactive tools (1)
- CPU/GPU Resource EstimatorSensor data throughput, memory and point-cloud load, with the assumptions and limits shown. It does not predict exact hardware performance.
E. Design challenges (1)
- Challenge 1: Design an Autonomous Mobile RobotIndoor differential-drive robot: sensors, motor sizing, compute, power, ROS 2, LiDAR+IMU, SLAM, Nav2, tuning and validation.
Practical deliverables
- Profiling procedures
- Benchmark examples
- Optimization checklist
Read first: existing FixTheRobot lessons and labs
This module builds on these instead of repeating them.
- Real-Time Performance and Resource Constraints SLAM · Lesson
- Evaluate Robustness and Latency Budgets Perception · Lab