Laying the foundations of a fully electric, autonomous wing-in-ground vehicle for zero-emission cargo transport across archipelagos, inland seas, and large lakes.


## Overview

Wing-in-ground (WIG) vehicles, also known as ekranoplans, fly a few metres above the water surface and exploit the ground effect to gain lift and reduce drag. This makes them more energy-efficient than conventional aircraft, with a smaller carbon and noise footprint, while being much faster than ships. They are particularly well suited to connecting islands, coastal regions, and communities around large lakes and inland seas.

Flying in ground effect is, however, difficult to control. Lift, drag, and altitude are strongly coupled close to the surface, waves and gusts continuously disturb the vehicle, and the transition between waterborne and airborne states is demanding. Because WIG craft have little operating history, there are also very few flight logs or validated aerodynamic models to build on.

**AIRSHIP** lays the foundations of a new class of fully electric unmanned aircraft system, the Unmanned WIG Vehicle (UWV). The project brings together:

- zero-emission electric propulsion combining solar and hydrogen sources with batteries and supercapacitors,
- advanced guidance, navigation, and control for fully autonomous flight in ground effect,
- on-board artificial intelligence for self-awareness and situational awareness, and
- environmental, economic, and social impact assessment together with a technology roadmap.

[![The AIRSHIP A-1 demonstrator with the project team](/projects/Airship/airship_vehicle.webp)](/projects/Airship/airship_vehicle.webp)

*The AIRSHIP A-1 technology demonstrator, with a 5.5 m wingspan, together with the project team.*

## Research Objectives

### Simulation and Digital Twins for Flight Control

SpaceR develops simulation tools and digital twins of the AIRSHIP vehicles, combining aerodynamic models with experimental data to approximate real operating conditions in ground effect. These environments provide the basis for designing, training, and testing flight control strategies before they are deployed on the prototypes.

### AI-Based Flight Control with Reinforcement Learning

Building on SpaceR's experience in modelling, simulating, and transferring deep reinforcement learning (RL) policies to real robotic systems. The project develops RL-based control policies for the different flight modes of the vehicle, including take-off, steady flight, manoeuvring, and landing.

### GNC and Perception Integration

SpaceR supports the integration of the guidance, navigation, and control (GNC) system, progressing from software-in-the-loop and hardware-in-the-loop testing to integration on the prototypes and field tests.

### Integration and Validation

The technologies are validated on a series of prototypes of increasing size: the A0-W, a 1:50 model for water channel and wind tunnel tests; the A0-S, a 1:10 flying model used to test aerodynamics and control strategies; and the A-1, a 1:5 scale technology demonstrator with a 5.5 m wingspan. In September 2026, the A-1 took off from water and completed its first short, stable flight in Tampere, Finland.

[![The AIRSHIP A-1 flying in ground effect over a lake](/projects/Airship/airship_shot_1.webp)](/projects/Airship/airship_shot_1.webp)

[![The AIRSHIP A-1 during a high-speed water run](/projects/Airship/airship_shot_2.webp)](/projects/Airship/airship_shot_2.webp)

[![The AIRSHIP A-1 during take-off from water](/projects/Airship/airship_shot_3.webp)](/projects/Airship/airship_shot_3.webp)

*The AIRSHIP A-1 demonstrator during water and flight tests in Finland.*

## Consortium

**Universidad Politécnica de Madrid (UPM) — Spain**  
Project coordination, model-based flight control, self-awareness for fault detection and reconfiguration, and zero-emission propulsion and power.

**Tampere University — Finland**  
Design of the AIRSHIP vehicle, including operational and technical requirements, and prototype flight testing.

**Tallinn University of Technology (TalTech) — Estonia**  
Economic, environmental, and social impact assessment.

**University of Luxembourg (SnT / SpaceR) — Luxembourg**  
Vehicle simulation and digital twins, reinforcement learning for flight control, and GNC integration.

**INESC TEC — Portugal**  
Perception and situational awareness, fusing radar, LiDAR, camera, and IMU data.

**La Palma Research Centre (LPRC) — Spain**  
Technology roadmapping, dissemination, and stakeholder engagement.

**Trisolaris Advanced Technologies — Portugal**  
Industrial partner.

## Online Material

- [Visit the AIRSHIP project website](https://airshipproject.eu/)
- [Browse the AIRSHIP project results on Zenodo](https://zenodo.org/communities/airship-euproject/records)


Autonomous Flying Ships for Inter-Island and Inland Waters Transport

Combining collaborative 3D mapping and extended reality to support mission planning and supervisory control of multi-robot exploration at the lunar south pole.


## Overview

Future lunar missions will deploy teams of robots to explore unknown and hazardous terrain, support in-situ resource utilization (ISRU) activities, and enable infrastructure development. Supervising these systems requires operators to maintain situational awareness across multiple robots in an unfamiliar environment. At the lunar south pole, extreme lighting conditions, hazardous terrain, and limited terrain observations make it difficult to fully assess the rovers' surroundings and plan operations from sparse images and separate robot data alone.

**ELYSIUM-XR** investigates how online 3D reconstruction can support mission planning and supervisory control of multi-robot systems for planetary exploration, with a direct focus on the lunar south pole. This PhD research aims to combine observations from multiple exploring robots into a shared, continuously updated 3D terrain representation. LiDAR-based reconstruction provides detailed terrain geometry under challenging illumination, allowing operators to inspect newly explored areas and adapt mission plans as information becomes available.

## Research Objectives

### Collaborative Mapping and Online Surface Reconstruction

The project brings together collaborative mapping and online surface reconstruction to combine observations from multiple exploring robots into a shared view of the terrain. This continuously updated representation is intended to help operators understand newly explored areas and identify hazards as the mission progresses.

### Interactive Mission Control in Extended Reality

An extended reality (XR) system will connect the reconstructed terrain with interactive mission control, helping operators understand their surroundings, plan operations, and coordinate robot activities during multi-robot exploration.

### Evaluation in Simulation and the LunaLab

The approach will be evaluated in simulation and in SpaceR's [LunaLab analogue facility](https://www.spacer.lu/facilities). Evaluation will consider reconstruction quality, system performance under realistic operational constraints, and effectiveness in supporting operators during multi-robot exploration.

## Project Information

- **Project number:** 18809546
- **Main supervisor:** Assoc. Prof. Miguel Angel Olivares Mendez
- **Researcher:** Alexandre Frantz
- **CET committee:** Miguel Angel Olivares Mendez, Malte Langosz, Carol Martinez Luna
- **Partners:** None

## Online Material

- [Watch the project video on YouTube](https://www.youtube.com/watch?v=WibrHY4XUO8)
- [Browse Alexandre Frantz's publications on Google Scholar](https://scholar.google.com/citations?user=tCgSZlQAAAAJ&hl=en)


Enhanced Lunar Mission Control System for Supervising and Controlling Unmanned Rovers with Extended Reality

Turning ice into thrust with solar energy — water electrolysis propulsion, in-space water extraction, and autonomous docking and propellant refilling for a self-sustaining space mobility infrastructure.



## Overview

The growing number of satellite launches, the emergence of in-orbit servicing (IOS) and the need for active debris removal (ADR) demand reliable, affordable and scalable solutions for in-space mobility.

**Ice2Thrust (S4I2T)** develops an innovative solar electric water propulsion system and builds around it a self-sustaining mobility infrastructure in space. Water is green, cheap, easy to store and transfer, and abundant on other celestial bodies, which makes it the core of the three technology pillars of the project:

- **Solar-electric water electrolysis propulsion** — an electrolyser decomposes water in flight into hydrogen and oxygen that feed hot-gas, cold-gas and electric thrusters from a common propellant supply.
- **Autonomous proximity operations, docking and propellant refilling** — enabling reliable, large-scale in-orbit servicing, debris removal and on-orbit assembly.
- **In-space water extraction and utilisation** — extracting water from icy regolith to refill spacecraft, closing the loop from ice to thrust.

## Research Activities at SpaceR

### Autonomous Proximity Technologies

SpaceR leads the development of learning-based guidance and control for close-proximity operations, docking and propellant transfer of small spacecraft. Reinforcement learning (RL) policies are trained to navigate a servicer to the docking point while satisfying thrust, attitude and safety constraints, such as approach corridors and range-dependent closing-velocity limits, and to keep the stack controlled while transferred propellant shifts its centre of mass and inertia.

### RL Training Environment for Docking and Refuelling

A massively parallel, GPU-accelerated six-degree-of-freedom simulation environment with orbital dynamics is developed to train and benchmark classical, model-free and model-based RL controllers for docking and refuelling scenarios across a range of spacecraft platforms. Additionally, a 6-DoF simulation environment for sim-to-real transfer of policies on the platforms and robots in Zero-G laboratory is also utilized.

### Docking and Refilling Hardware-in-the-Loop Demonstration

Trained policies will be transferred zero-shot to the floating platforms of SpaceR's Zero-G Lab, with a goal of hardware-in-the-loop demonstration of autonomous in-orbit servicing operations.

## Consortium

**Technical University of Munich (TUM)**
Project coordination, water extraction and water electrolysis propulsion.

**University of Luxembourg — SnT / SpaceR**
Autonomous proximity technologies, RL training environment, and docking and refilling hardware-in-the-loop demonstration.

**EnduroSat**
Spacecraft technologies and CubeSat mission analysis.

**TechTour**
Communication, dissemination, market analysis and commercialisation.

## Publications

- A. Arora, R. Marsal i Castan, M. El-Hariry, and M. Olivares-Mendez. "SCoCaT: Success Conditioned Constrained Reinforcement Learning for Spacecraft Docking." *Accepted at CoRL 2026.* [arXiv](https://arxiv.org/abs/2609.06061)
- A. Arora, P. Lima, and M. Olivares-Mendez. "Massively GPU-Accelerated Reinforcement Learning with Constraints for Autonomous Spacecraft Docking and Fuel-Transfer Alignment." *Presented at the ESA GNC & ICATT Conference 2026.*


Solar for Ice to Thrust

Enabling Earth-to-Moon robotic assembly despite multi-second communication delays by combining extended reality, delay-robust haptic control, and robot skill learning.


## Overview

Future lunar bases will rely on robots to assemble and maintain infrastructure, with human operators on Earth staying in the loop for key decisions. The Earth–Moon ~3s round-trip delay makes direct teleoperation unstable and mentally demanding, especially when the operator receives force feedback. At the same time, full autonomy cannot yet handle every situation in a partially known lunar environment.

**LunarAssembly** addresses this by combining immersive Extended Reality (XR), delay-robust haptic control, and robot skill learning in a shared autonomy framework. The robot learns assembly skills from human demonstrations and executes them on the remote site, while the operator can:

- preview and adjust a skill in XR before it is executed,
- take over and correct the robot at any moment during autonomous execution,
- fall back to direct force-feedback teleoperation that remains stable despite the delay, and
- perceive the remote scene as a 3D reconstruction with detected hazards, rather than only through 2D camera feeds.

[![LunarAssembly framework overview](/projects/LunarAssembly/LunarAssembly_whitebg.webp)](/projects/LunarAssembly/LunarAssembly_whitebg.webp)

*Overview of the LunarAssembly framework, integrating teleoperation, shared autonomy, skill learning and XR-based teleoperation.*

## Research Objectives

### Immersive XR Telepresence

The remote scene is reconstructed in 3D from the robot's LiDAR data and rendered in XR, giving the operator a navigable view of the work area. Rocks and craters are segmented using a deep learning model and displayed as 3D hazards, helping the operator cope with the low sun angles and long shadows expected at the lunar south pole.

### Delay-Robust Haptic Teleoperation

LunarAssembly builds on the Time Domain Passivity Approach to keep force-feedback teleoperation stable under multi-second delays. The project extends it so that the required damping is spread over time and separated from the operator's own commands, so stability does not come at the cost of operator comfort.

### Skill Learning and Shared Autonomy

Assembly skills such as picking, placing, and pressing parts into place are taught by demonstration in a delay-free lab setting and encoded as task-parameterized models that generalize to new object positions. On the remote site, these skills run autonomously, and the operator's corrections can be used to refine them incrementally.

### Integration and Validation

The components are being integrated and tested in SpaceR's LunaLab, a lunar analogue facility with a basalt test bed and controllable lighting. A mobile manipulator has already been teleoperated between TU Wien in Vienna and the LunaLab in Luxembourg (~760 km apart) over the public internet. The next steps are to develop the XR framework with 3D reconstruction, integrate it with the teleoperation system, and begin testing under emulated Earth–Moon delays.

## Consortium

**University of Luxembourg — SnT / SpaceR**  
Project coordination, XR framework with 3D reconstruction and AI-based hazard detection, system integration, and experimental validation in the LunaLab.

**TU Wien — Robotic Systems Lab**  
Delay-robust haptic control and passivity-based force-feedback teleoperation.

**TU Wien — Autonomous Systems Lab**  
Robot skill learning from demonstration and shared autonomy.

## Publications

- H. T. Dinc, J. Heidersberger, C. Kyriazis, T. Wagner, A. Frantz, D. Pisanti, F. Kong, M. A. Olivares-Mendez, D. Lee, and C. Ott. "Earth to Moon Bilateral Teleoperation for Lunar Assembly: A Unified Framework for Shared Autonomy, Skill Learning, and Telepresence." *Accepted at the 2026 IEEE Conference on Telepresence.*


Large Time-Delay Teleoperation for Enabling Robotic Assembly on the Moon through Extended Reality and Machine Learning

Developing robust and autonomous heterogeneous robotic systems that can anticipate failures, adapt to challenging lunar terrain, and collaborate efficiently during future lunar exploration missions.


## Overview

Future lunar exploration will increasingly rely on autonomous robots to operate in terrain and environmental conditions that are difficult or impossible for humans to access directly. Current robotic systems, however, still have limited autonomy when dealing with challenging terrain, unexpected conditions, and failures.

**NeverGiveUp** develops a robust and highly autonomous framework for heterogeneous multi-robot lunar exploration. The project brings together wheeled and legged robots with complementary locomotion, manipulation, sensing, and planning capabilities.

The central idea is to enable robotic teams to:

- identify and predict potential failures,
- react before significant damage occurs,
- define, share, and execute tasks collaboratively, and
- adapt their behaviour to local terrain conditions and robot state.

By combining failure-aware autonomy, global mission planning, robust robotic skills, and multi-robot collaboration, NeverGiveUp aims to make future lunar prospection missions more efficient, resilient, and sustainable.

![NeverGiveUp multi-robot mission planning concept](/projects/NeverGiveUp/ngu-planning-framework.png)

*Exemplary view of the NeverGiveUp planning framework, where multiple paths are planned for heterogeneous robots and local environmental information is continuously fused into the global planner to support frequent replanning.*

## Research Objectives

### Failure Detection and Damage-Minimizing Reaction

NeverGiveUp develops AI-based methods to detect, identify, and predict internal robot failures and mobility risks using onboard proprioceptive sensing.

For wheeled robots, this includes situations such as excessive wheel slip, sinkage, entrapment, and degraded mobility. For legged robots and manipulators, the framework also considers internal failures and loss of performance.

The project further investigates reinforcement-learning-based recovery strategies that allow robots to react to critical situations before they result in significant damage or mission failure.

### Autonomous Global Mission Planning

The project develops a global mission-planning framework for heterogeneous robotic teams.

The planner combines prior information such as satellite maps with information collected locally by the robots, including terrain properties, scientific targets, robot capabilities, and robot health.

As new information becomes available or robot capabilities change, the mission can be replanned dynamically and tasks redistributed among the robotic team.

### Robust Robotic Skills

NeverGiveUp develops robust locomotion and manipulation skills for planetary exploration.

These include sample collection, scientific instrument deployment, object and regolith manipulation, and collaborative tasks that require multiple robots.

The objective is not only to enable these skills under nominal conditions, but also to maintain useful operation under challenging terrain and partial robot failures through adaptive fallback strategies.

### Integration and Validation

The different components of NeverGiveUp will ultimately be integrated and evaluated as a complete heterogeneous multi-robot system.

Experimental validation will be performed in lunar analogue environments, including SpaceR's LunaLab at the University of Luxembourg, together with outdoor analogue testing.

The final demonstrations will combine failure-aware autonomy, mission replanning, robust robotic skills, and collaborative exploration.

## Consortium

**University of Luxembourg — SnT / SpaceR**  
Project coordination, AI-based failure detection and prediction, reinforcement-learning-based recovery, system integration, and experimental validation.

**ETH Zurich — Robotic Systems Lab**  
Global mission planning, collaborative exploration, legged robotics, and robust robotic skills for planetary exploration.

## Publications

- R. Viviano, S. Omi, A. Orsula, and M. Olivares-Mendez. "Fleet-To-Lab: A Transfer Learning Framework For Lunar Rover Slippage Estimation Via Model Fusion." *Accepted for presentation at [iSpaRo 2026](https://isairas-isparo.de/).* [arXiv](https://arxiv.org/abs/2609.17187)

Fallback Adaptable Heterogeneous Multi-Robots for Efficient and Collaborative Lunar Prospection

Advancing real-world reinforcement learning for autonomous robots in unstructured environments, with applications to floating waste collection with surface vehicles and space debris capture with orbital robots.


## Overview

In the last decade, Reinforcement Learning (RL) has shown exceptional results on a wide range of tasks, in particular when applied to video games. Most strikingly, these agents have been able to solve both games that require fast, nimble actions, and deep, well-planned decisions. When applied to robotics, these agents have shown similar aptitudes, often outperforming their classical control counterparts. The recent progress in robotic manipulation and legged robot locomotion is a representative example of their capacities on low-dimension continuous control tasks.

The advent of RL, Neural Networks (NN), and ever-more-efficient edge devices are offering the ability to deploy high-performance visuomotor policies in real time on robots. However, key challenges remain: simulators capable of recreating high-fidelity environments are slow, and their visual appearance remains largely different from the real world — particularly for outdoor applications where environmental complexity is much higher. This leads to a gap between what can be simulated and the deployment environment.

The main developments in the R3AMA project are focused on RL solutions applicable to real robotic systems operating in unstructured environments with real sensors. RL is applied to sensorimotor tasks, coordination of tasks on a single agent, and coordination between agents in a shared environment. The core tools are model-based reinforcement learning and high-performance, high-fidelity simulation. The project targets space applications, including the automation of space debris capture and de-orbiting with robotic satellites operating in orbit.

## Research Activities

### Multi-Task Reinforcement Learning for Space Robotics

A core challenge in deploying autonomous agents in space is the need to perform multiple distinct tasks — navigation, inspection, proximity operations, and manipulation — within a single unified policy. R3AMA develops RL frameworks that learn shared representations across tasks, enabling agents to transfer knowledge between objectives and adapt to novel mission requirements without retraining from scratch.

### Sim-to-Real Transfer

Orbital environments present extreme sim-to-real challenges: zero-gravity dynamics, harsh lighting conditions, and the absence of GPS or standard sensing modalities. To bridge this gap, R3AMA employs domain randomisation, physics-informed simulation, and high-fidelity rendering to train policies that generalise robustly to real hardware.

### Autonomous Navigation

Autonomous agents operating in proximity to other spacecraft must navigate safely while avoiding collisions and respecting fuel constraints. R3AMA applies deep RL to develop robust navigation policies that account for disturbances, partial observability, and the unique dynamics of orbital mechanics.

## Publications

- El-Hariry, M., Richard, A., Marsal, R., Batista, L. F. W., Geist, M., Pradalier, C., & Olivares-Mendez, M. (2025). RoboRAN: A Unified Robotics Framework for Reinforcement Learning-Based Autonomous Navigation. *Transactions on Machine Learning Research*. [Read](https://openreview.net/forum?id=0wDbhLeMj9)

## Partners

**CNRS IRL2958 GT-CNRS** — France  
Joint research lab of CNRS and the Georgia Institute of Technology. Expertise: field robotics, AI, robotics for natural environments.  
*Cédric Pradalier, Stéphanie Aravecchia*

**IADYS** — France  
SME commercialising Unmanned Surface Vessels (USV) for waterway cleaning. Expertise: USVs, system integration, computer vision.  
*Nicolas Carlesi, Ronald Loschmann*

**University of Luxembourg — SpaceR** — Luxembourg  
Space Robotics laboratory. Expertise: space robotics, AI, reinforcement learning, Zero-G environments.  
*Miguel Olivares-Mendez, Antoine Richard, Ricard Marsal i Castan*


Exploring the Frontiers of Space Robotics

Research Themes

Planetary Robotics

Orbital Robotics

Space Manipulation

Research Facilities

LunaLab

Zero-G Lab

Latest Projects

Autonomous Flying Ships for Inter-Island and Inland Waters Transport

Enhanced Lunar Mission Control System for Supervising and Controlling Unmanned Rovers with Extended Reality

Solar for Ice to Thrust

SpaceR Team