Swiss Aerospace Ventures
All articles

JARVIS4Pre: teaching robots to drape carbon fibre with reinforcement learning

Take Off Insights 2026 · Session "UAS & Testing" · Ulrich Kuenzer, Alpex Technologies (Mils, Tyrol)

Julian Walder·
Swiss Aerospace Ventures summary slide: JARVIS4Pre: teaching robots to drape carbon fibre with reinforcement learning
Key figures at a glance. Summary slide: Swiss Aerospace Ventures, based on the talk at Take Off Insights 2026.

Preforming, the step where flat dry carbon fabric is shaped into a three-dimensional part before resin infusion, is still largely a craft. Multiple plies have to follow a complex geometry without wrinkles or defects, and the knowledge sits in the hands of experienced operators. JARVIS4Pre set out to capture that knowledge in software.

At a glance

Consortium Alpex Technologies (lead, physical validation), AAC aerospace & advanced composites (kinematic draping and infiltration simulation), enliteAI (surrogate model, AI work instructions), FH Upper Austria (material characterisation, digital twin, FEM), BRIMATECH (innovation environment, exploitation)
Programme FFG Take Off, call 2020
Goal More automation, repeatability and quality assurance in complex aviation preforms

The approach

  1. Build a surrogate model of the draping process, so simulations run fast enough to train AI.
  2. Create a digital twin and a scalable training environment for AI agents.
  3. Train reinforcement-learning agents on the surrogate to generate work instructions.
  4. Validate with finite-element simulations and a physical robot cell.

Results

The team delivered the full chain: material characterisation, an FE draping model, the surrogate model, agent training, optimised draping trajectories and physical validation in Alpex's robot cell. The work is published as "Towards the automation of woven fabric draping via reinforcement learning and Extended Position Based Dynamics" in the Journal of Manufacturing Processes (2025).

What's next

The framework will be extended to complex geometries, thick stacks and different fabrics. The longer-term goal is a fully automated chain: 3D scanning, then cloth simulation, then reinforcement learning, then robotic control, for arbitrary tool geometries. Follow-ups include AutoDrape (Upper Austrian doctoral programme), Pre-2-K (FFG, draping simulation for diaphragm forming) and a submitted Josef Ressel Centre proposal for an as-built simulation platform for composites.

Related: ARIADNE: AI surrogate models for aero-engine design (coming on 22 Nov) · AeroDOM: distortion-optimised titanium machining (coming on 21 Nov)

Back to the overview: Take Off Insights 2026

People and organisations

  • Ulrich Kuenzer, Speaker, Alpex Technologies

Organisations: Alpex Technologies, AAC aerospace & advanced composites, enliteAI, FH Oberösterreich, BRIMATECH

Sources

  • Talk and slides by Ulrich Kuenzer, Take Off Insights 2026, Vienna, 12 November 2026

Share this article