Real-Time Convex Optimization
summary
Claims that convex problems (especially QPs) run in 50 microseconds on small embedded hardware and that automatic compilation to cone programs plus code generation (example: CVXGen) yields 1,000x speedup over generic solvers. The pipeline emits flat C (library-free) solvers, fits real-time constraints, trades off footprint and reliability, handles warm starts and fixed sparsity known at compile time, and can target FPGAs for much lower energy even if latency gains vary.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:32Convex Optimization
- 1:51Why Convex
- 2:56State of the art
- 3:20Domainspecific languages
- 3:50Rapid prototyping
- 4:20Support Vector Machine
- 5:57RealTime Embedded Optimization
- 7:14RealTime Convex Optimization
- 9:58Example
- 11:02What do you need
- 12:06General solver
- 13:27parser solver
- 14:12CVXGen
- 16:44Conclusion
- 19:10Missing Features
related talks
39mRobotics at Compile Time: Optimizing Robotics Algorithms With C++'s Compile-Time Features
01h 03mOLD VERSION: Building Abstractions at the Hardware-software Boundary
43mSoftware Development Trends for FPGAs
28mAccelerating Performance of Qt on TI's Embedded Processors
30mCFU Playground: Model-specific Acceleration on FPGAs
24mUniversal Tools for Acceleration, Timing, Integration & Machine Enhancement
13mPReGO: A Generative Methodology for Satisfying Real-Time Requirements on COTS-Based Systems: Definit
15mCertified and Efficient Instruction Scheduling: Application to Interlocked VLIW Processors