Skip to main content
Back to News

PhD Opening: AI-Assisted Design of Deformable Robot Digital Twins (PEPR O2R / AS4)

May 13, 2026
· 3 min read
Research
pepr-o2r soft-robotics phd-opening sofa ai

I am recruiting a PhD student for a 3-year funded position at the intersection of generative AI and deformable robotics simulation.

Project Overview

Creating a digital twin of a soft robot in SOFA/SoftRobots today requires deep expertise in geometric modeling, continuum mechanics, C++/Python scene programming, and FEM solver configuration — a barrier that slows design iteration considerably. This thesis aims to break that barrier.

The core objective: develop a multi-agent AI framework (LLMs + specialized tools) that translates a plain-language description of a deformable robot —

“a pneumatic silicone gripper, Young’s modulus 1 MPa, grasping 500 g objects with two variable-pressure actuators and deformation sensors at the tips”

— into a complete, simulation-ready SOFA/SoftRobot digital twin.

Structured Missions (3 Years)

  1. State of the art & benchmarking (months 1–6) — Survey LLM/agent approaches for physical model generation (CAD, FEM, simulation); build a benchmark of NL descriptions → SOFA scenes covering geometry, materials, and actuators.

  2. Multi-agent framework design (months 7–18) — Hierarchical architecture with specialized agents: planner, geometry generator (VLM + mesh generation), materials & physics agent (RAG on constitutive law databases and metamaterials), actuator/sensor placement agent, and SOFA Python/C++ code generator. Verification loop via simulation execution and RL/gradient-based optimization.

  3. Validation & case studies (months 19–30) — Real soft robot applications (grippers, manipulators, hybrid rigid–soft locomotors). Quantitative evaluation of twin fidelity, design time reduction, inverse control precision, and sim-to-real transfer.

  4. Dissemination & open source (months 6–36) — Publications at ICRA, IROS, RoboSoft, IEEE T-RO; release of an AI-SoftTwin SOFA plugin; contributions to AS4 deliverables of PEPR O2R.

Funding & Collaborations

The position is funded by Action Structurante AS4 — “Modeling, Simulation, Multi-scale and Biomechanics” of the PEPR O2R (Programme de recherche sur la robotique organique), a national programme involving Inria, CEA, University Paris 1, ENSMM, Institut Mines Telecom, and others.

A parallel PhD in Humanities and Social Sciences will run alongside this engineering thesis (socio-anthropological dimensions of digital twins in soft robotics). Both candidates will co-organize workshops and may co-publish, reflecting PEPR O2R’s commitment to interdisciplinary research.

Candidate Profile

  • Strong foundations in robotics (deformable/soft robotics or control preferred), continuum mechanics, and numerical simulation
  • Python fluency required; C++ a strong plus; experience with SOFA, MuJoCo, PyBullet, or any FEM tool highly valued
  • Modern AI knowledge: LLMs, advanced prompting, multi-agent systems, RAG, fine-tuning (PyTorch / Hugging Face)
  • Master 2 in Robotics, Computer Science, Mechanics, or AI
  • Open-source contributions or publications: a plus

How to Apply

Send your CV, transcript, and a brief motivation letter to .