Synthesis of Controllers for Continuous Blackbox Systems
summary
Method synthesizes a continuous feed-forward controller for a stateless blackbox plant by building a linear interpolation model from a few hundred guided samples, using Delaunay triangulation to propose candidate control functions and verifying them under a known Lipschitz bound. Heuristics pick high-uncertainty samples to break ambiguous triangles and falsify spurious candidates; final verification uses extra samples to give sound guarantees. Evaluation includes random benchmarks and a transistor-level band‑gap regulator where other methods fail and the approach finds solutions in ~150–300 simulations.
AI-written summary · about
chapters (12)
- 0:00Blackbox controller synthesis intro
- 1:04Problem setting and system
- 3:13Synthesis approach overview
- 5:04Modeling with Delaunay triangulations
- 8:43Exploratory sampling heuristics
- 13:40Finding candidate control functions
- 14:47Verification and model refinement
- 16:13Formal guarantees
- 18:00Experimental results
- 21:40Industrial case study: voltage regulator
- 22:24Conclusion and future
- 22:45Q&A