Rene-Marcel Lehner
Rene-Marcel Lehner

Projects

Being expanded: more projects, fuller case studies, and documents are being added.

The assembled follower arm, a printed six-joint arm with a two-finger gripper
2026 – present
Hardware assembled

Recover in Real Time

A robotic-arm platform, assembled and calibrated, for recovery under delay.

The physical endpoint of my research programme: a two-arm leader–follower platform, assembled and calibrated, on which the planned recovery-and-delay study will run. Everything shown today is assembly, calibration, and teleoperation, explicitly not autonomous execution.

Assembled and calibrated the leader–follower arm pair, built the teleoperation and diagnostics workflow, and designed the planned 2×2 study of recovery data and delay-aware execution.

Repository

Read the case study

Two-link arm in its true world, with the nominal model’s prediction from a moment earlier drawn faded; the gap between the two hands is the residual
2026 – present
Active research, results pending

Residual Worlds

Keep the known physics, learn only the mismatch, then judge it inside MPC.

A controlled test of a simple principle: keep the known physics, learn only the residual, and evaluate the learned dynamics model by closed-loop control performance rather than prediction error. Simulator and protocol are being implemented; calibration and protected evaluation are pending.

I am designing the study, implementing the simulator and the MPC stack, and preregistering the evaluation protocol before any model is trained.

WebsiteRepository

Read the case study

Two policies running the same symbolic gridworld scenario side by side
2026
Release preparation

Grounded Recovery

When expert time is fixed, does labeling the learner’s recoverable mistakes buy more robustness than collecting more nominal behavior?

Release preparation: public repository, report, and evidence bundle have not yet passed publication review.

WebsitePreliminary report (36p, 1.6 MB)Repository

Two dotted lattices turning against each other, their overlap forming a shifting moiré pattern
2025 – present
Research continuation

Envelope Approximation for Photonic Moiré Crystals

A two-scale theory for twisted photonic crystals, rebuilt after its own audit.

My M.Sc. research developed a multiband two-scale envelope theory for twisted photonic crystals. A post-thesis audit exposed invalid comparisons in the submitted validation; I am rebuilding the model and evaluation against independently converged references.

I derived the two-scale theory, implemented the research pipeline behind it, and am now conducting the post-thesis audit and rebuild. The work was my supervised M.Sc. research at TU Dortmund University.

Thesis (92p, 28.1 MB)Manuscript (25p, 2.7 MB)Defense slides (12.4 MB)MSL framework source

Read the case study

The Blaze2D mark
A photonic band diagram of a hexagonal lattice of air holes in a dielectric, drawn band by band, with the complete band gap shaded between the third and fourth bands
2025 – 2026
Validated release

Blaze2D

A Rust Maxwell eigensolver, validated against MPB, in the browser and in Python.

An independently developed 2D Maxwell eigensolver for inspectable, high-throughput Bloch-parameter studies, validated against MPB and exposed through Rust, Python, and WebAssembly.

I designed and implemented the solver independently: the numerical formulation, the mixed-precision eigensolver core, the benchmark and validation suite, and the Rust, Python, CLI, and WebAssembly interfaces.

WebsiteTechnical reportRepositoryPyPIReport (16p, 0.7 MB)Manuscript (12p, 0.9 MB)

Read the case study

Particles aligning into flocks under the sampled neighbor rule
2023
Archived

Neural Swarm Dynamics

Sampling a few neighbors reproduced flocking; the RL extension did not.

Bachelor-thesis research: a stochastic metric-topological neighbor rule approximates the Vicsek alignment vector and reproduced its order-parameter behavior, which is the thesis contribution. A separate multi-agent RL extension ran end to end but did not converge; both outcomes are reported.

I implemented the C++/OpenMP simulation and its Python bindings, the sampling analysis, and the actor–critic multi-agent RL pipeline, and wrote the thesis and manuscript.

Thesis (37p, 5.4 MB)Manuscript (8p, 2.9 MB)Defense slides (6.8 MB)Repository

Read the case study