Learn Robotics with C++ in 1 Hour
summary
Claims that localization needs multiplying tiny probabilities across a laser scan and that a single scan can be on the order of 900000 points leads to arithmetic underflow unless code uses log space or special handling. Benchmarks show some safe numeric library (no float support) throw-on-overflow designs are unusable on the hot path. Perception stacks use OpenCV PCL and deep frameworks like Caffe TensorFlow or DIB and rely on GPUs; practical systems aim for ~10 Hz and sometimes run on NVIDIA TX1 with 16‑bit floats for performance.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:51Robotics and challenges
- 15:58Perception and object ID
- 23:21Deep learning in robotics
- 30:13GPUs and CUDA for robotics
- 38:02Cross-process communication
- 41:16Middleware frameworks
- 47:11Testing and debugging
- 51:52Future of C++ in robotics
- 55:47Q&A
related talks
01h 26mC++ on Mars: Incorporating C++ into Mars Rover Flight Software
37mArchitecting Multithreaded Robotics Applications in C++
43mLightweight, low-overhead, high-performance: machine learning directly in C++
01h 22mCostless Software Abstractions for Parallel Architectures
20m"Rust in Motion: Building Reliable and Performant Robotics Systems"
30mLost in translation?! Robotics with type-safe coordinate systems
35mBreaking language barriers:High-performance IPC with iceoryx2
01h 05mThe Most Important Optimizations to Apply in Your C++ Programs