EST. 2021 Antwerp, Brussels, Leuven, Liège & Remote Now talking · Q1 2027

Research, software, and aerial robotics.

Peinser is a Belgian engineering team. Applied research, platform engineering, and aerial robotics, typically two of the three on any given engagement.

Start a conversation
01 — Practice

Research & development.

Applied research, machine learning, and engineering R&D. We take open questions, yours or ours, and produce reproducible prototypes, peer-reviewable results, code and hardwarethat runs in production.

Applied ML · Data & modelling · Lit reviews · Co-publications
02 — Practice

Platform engineering.

Distributed systems, cloud infrastructure, CI/CD, and the observability layer underneath. We design platforms, harden DevOps practice, and embed with your team to move systems from prototype to production.

Distributed Systems · IaC · Kubernetes · CI/CD · SRE & Observability
03 — Practice

Aerial robotics.

Mission planning, swarm coordination, and ground-control software for autonomous UAV fleets. The newest part of the practice, and the fastest growing.

Swarm control · Active-control lock · ML in the loop · BVLOS Ops
04 — Practice

Strategy & direction.

Technical audits, discovery, and decision memos for teams that need a second opinion before they commit. The deliverable is the document; the value is the choice it forces.

Technical audits · Decision memos · Roadmaps · Second opinions
— 01 / Robotics

Aerial robotics, end to end.

Most of our robotics work is invisible; coordination, deconfliction, data streams and machine learning running quietly underneath. The operator sees one map, one mission, one outcome.

— Practice 03 / Aerial robotics

Technology for UAV fleets.

We build the layer between operator and swarm: turning a brief like "inspect this 14-hectare site" into coordinated flight plans: geofenced, deconflicted, and logged. We design the airframes too, optimised for your software stack.

  • Active control lockA cryptographic authority model, only the on-shift pilot can issue commands; hand-offs are signed, audited, and reversible.
  • Machine learning in the loopOn-device perception (NVIDIA Jetson), anomaly detection, and target re-identification. Eight years of production ML, applied at the edge of the airframe.
  • Swarm coordinationDeconfliction, formation control, and leader–follower behaviour.
  • Detect & avoidADS-B, detection networks, and onboard sensors fused into a single airspace picture.
  • Ground-station UXOne operator, many drones. Mission planners with an explicit authority model and inline regulatory checks. Secure communications, control, and telemetry with Tor, MeshCore & WireGuard.
  • BVLOS readinessConOps, audit trails, and the regulatory work that makes beyond-visual-line-of-sight operations legal in the EU.
  • Autonomy integrationROS 2, PX4, ArduPilot, Betaflight, OpenPilot, custom MAVLink or MSP implementation. We plug into the stack you already fly, we don't replace it.
— 02 / Method

A six-week engagement shape.

Long enough to think rigorously. Short enough that the result still matters. Every engagement ends with something a team can act on. A working prototype, a decision memo, or a flight plan.

01 — Phase

Discovery.

Read the literature, the code, the data, and the runbooks. Talk to operators and end users. The deliverable is a findings memo with explicit assumptions and open questions.

WK 1–2 · Findings memo
02 — Phase

Decision.

A small set of options, each with a cost, a risk profile, and an owner. We choose together; the choice — and the rationale — is the deliverable.

WK 3–4 · Decision memo
03 — Phase

Ship.

Two weeks to put the decision in front of real users — or in the air. Not a demo: a thin, instrumented slice of the real system, deployed and observable.

WK 5–6 · v0.1 in production
The hardest engineering decisions are rarely about what to build. They are about what to keep running, what to deprecate, and which version is worth the cost of operating. — Internal note · Wk 03 · Aerial inspections engagement
— 03 / Projects

Products and platforms.

A handful of products and platforms we keep building between engagements. All of them feeding back into the work we do for clients.

Product · Documents Live · v1.0

Xarta

A document protocol for the full lifecycle: render, sign, archive, dispatch.

Every project ends up reinventing the same plumbing: rendering PDFs from templates, versioning the templates, signing the output, zipping bundles, attaching them to email, dispatching Peppol invoices, archiving for the legal retention period. Each team rolls its own subset, badly.

Xarta consolidates the lot behind a DAG protocol. The submitter declares the flow; Xarta executes it deterministically. Plug-in template engines, templates versioned in git rather than locked in a vendor UI, Peppol BIS 3.0 in the same pipeline as everything else. On-prem by default.

On-prem · Peppol BIS 3.0 · Document Lifecycle Template Rendering · PDF/A · PAdES
Open source · Geodesy Live · v0.8

CORSHub

A federated CORS / RTK network for centimetre-grade GNSS.

An auditable community network of base stations that broadcasts RTCM corrections to anything with an RTK receiver; surveyors, agricultural rovers, and our own drones. Open protocol, open data, federated by design.

github.com/peinser/corshub MIT · NTRIPv2 · Open Policy Agent·
Project site →
Hardware · Catalog Shipping

Airframes

A small catalogue of drones, built to fly with our software.

Three reference airframes; a long-range FPV quad, a long-endurance fixed-wing, and a heavy-lift inspection platform. Open BOMs, open firmware, optimised for swarm flight against our ground station. Buy one, or build your own from the spec sheet.

3 airframes · €4.2k–€18k ArduPilot · Betaflight · PX4 · OpenPilot · OpenFOAM CFD
Spin-out · Energy · EV Charging In market

Zonnetanken

Smart EV charging that follows the sun.

An open charging network of EV chargers with auditable charging sessions for employee cost recovery. Started as a side project; now an independent product with paying users across Belgium. Completely exposes the internal chargepoint API through OCPP for complete ownership of your charging infrastructure. This allows users to maintain full control over their charging data and operations.

zonnetanken.be Web · OCPP · Alfen · Peblar
Visit Zonnetanken →
— Selected clients
Acerta ULiège Innovatix Duvel Logistics H.Essers Van Cleven NV
— 04 / Research

Research and open source.

Selected publications and code we have shared with the community. Most of this work feeds back into client engagements.

Research paper · ICML 2021

Likelihood-free MCMC with amortized approximate ratio estimators.

  • Machine learning
  • Simulation-based inference
  • MCMC
proceedings.mlr.press ↗ Research paper · NeurIPS 2022

Towards reliable simulation-based inference with balanced neural ratio estimation.

  • Machine learning
  • Bayesian inference
  • NeurIPS
papers.nips.cc ↗ Research paper · TMLR 2022

A crisis in simulation-based inference? Beware, your posterior approximations can be unfaithful.

  • Machine learning
  • Simulation-based inference
  • Calibration
openreview.net ↗ Research paper · MNRAS 2022

Constraining warm dark matter with stellar streams and neural networks.

  • Machine learning
  • Bayesian inference
  • Astronomy
academic.oup.com ↗ Code repository · amortized-experimental-design

Learning to select optimal experiments.

  • Statistics
  • Machine learning
  • Tutorial
  • PyTorch
github.com/JoeriHermans ↗
Now talking · Q1 2027

Let's talk about your problem.

If you have a research question, a system that needs building, or a fleet to manage — meet with us. The conversation is free, and we will tell you how we would approach it.

Start a conversation