Nonlinear control discovery
Explore interpretable adaptive, robust, and model-based strategies beyond textbook examples.
RESEARCH PROGRAM / 02 · CONTROL DISCOVERY
Discover control.
Test its limits.
An open research program using autonomous AI research agents to discover, falsify, and refine control strategies for uncertain nonlinear systems.
FOUNDING PHASE · OPEN RESEARCH ROADMAP
FIG. 02 / CONTROL DISCOVERYCONCEPT ILLUSTRATION / NOT A MEASURED RESULT
RESEARCH DOMAIN
Autonomous control research
CURRENT STAGE
01 / FOUNDATION
METHOD
Hypothesis → experiment → evidence
THE MISSION
Real engineering. Measurable evidence.
Build a reproducible discovery engine for control laws that are more robust, efficient, or capable than established baselines.
Can AI discover a controller that outperforms a strong, fairly tuned baseline under unfamiliar conditions, and reproduce the result?
We treat AI-generated engineering ideas as hypotheses, not accomplishments. The research loop demands explicit assumptions, independent tests, reproducible artifacts, and comparison with strong conventional methods. No performance claim is published without a traceable experiment.
THE RESEARCH AGENDAThree frontiers.
Three frontiers.
One rigorous method.
A focused technical agenda, structured to produce useful open research artifacts rather than speculative demonstrations.
Automated falsification
Use randomized environments, adversarial disturbances, and carefully separated test sets.
Efficient implementation
Measure performance, inference cost, and latency with eventual embedded deployment in mind.
How each research cycle works
HYPOTHESIZE
Explore classic, symbolic, learned, and hybrid control approaches.
SIMULATE
Run controlled experiments across disturbances, uncertainty, and system dynamics.
CHALLENGE
Evaluate candidates on held-out conditions and independent metrics.
PROMOTE
Advance only improvements supported by repeatable experimental evidence.
The proving ground. For aircraft problems, the plan is to challenge candidates with the flight tools: nonlinear simulation with turbulence, Monte Carlo dispersions, and MIL-F-8785C flying-qualities criteria, run independently of the agents that proposed them.
THE ROADMAP
From first principles to demonstrated capability.
Milestones are intentions, not achieved results. Progress and benchmark evidence will be added as they become independently verifiable.
I / FOUNDATIONCURRENT FOCUS
FOUNDATION
Deterministic plant models and baseline controller comparisons
II / SEARCHPLANNED
SEARCH
Autonomous candidate generation with independent validation
III / ROBUSTNESSFUTURE
ROBUSTNESS
Cross-system benchmarks and rigorous holdout testing
IV / HARDWARELONG-TERM
HARDWARE
Deploy verified controllers on dedicated compute hardware