Lots of interest for our poster on RGB and thermal scene reconstruction. I’m glad people were as surprised about our results as I was originally :).

#icra #robotics #computervision #science #machinelearning #artificialintelligence #multimodal

Travelling to #ICRA2026 this weekend?

Get ready for the journey by downloading some episodes of Robot Talk! There's over 150 episodes in the back catalogue, covering everything from soft robotics to industrial robots.

Available wherever you get your podcasts: https://linktr.ee/robottalkpod

#Robots #Robotics #ICRA

We are headed to #ICRA 2026!

We will be showcasing two of our latest research breakthroughs at the International Conference on Robotics and Automation (ICRA) next week.
_

1️⃣ D-CAT 😼: Decoupled Cross-modal Knowledge Transfer
📍 Track: Interactive Session 6 − Hall C (Poster Session)
📅 Date/Time: Thursday, 15:00-16:30, Paper ThI2I.59
🧑‍🔬 Authors: Leen Daher,* Zhaobo Wang, Malcolm Mielle

Our work on D-CAT solves a major real-world challenge: training a system with rich, multi-modal data (like video, IMU, and audio) but allowing it to operate using only a single sensor during inference.

By enabling cross-modal knowledge transfer, D-CAT reduces hardware redundancy and costs without sacrificing accuracy—even showing up to a 10% F1-score gain in certain scenarios!

Paper: https://arxiv.org/pdf/2509.09747

* Leen Daher is currently searching for a PhD/job in the EU, and will be presenting in person.
_

2️⃣ Unpaired Multi-Modal Reconstruction 📸🔥
📍 Track: Workshop Paper - MM-SpatialAI Workshop
📅 Date/Time: June 1st
🧑‍🔬 Authors: Jean Cordonnier, Chenghao Xu, Olga Fink, Malcolm Mielle

We introduces a framework for independent RGB-Thermal Novel View Synthesis reconstruction. We’ve developed a framework combining VGGT, feature matching, and multi-modal 3DGS to accept decoupled RGB/TIR image sets, proving that misalignment between modalities doesn't have to compromise the quality of the reconstruction.

#ICRA2026 #Robotics #MachineLearning #ComputerVision #MultiModal #SensorFusion

Ok so D-CAT 😼 was accepted to #ICRA! Train with lots of sensors, deploy with much less!

If you want to use cross modal transfer learning for when your sensor stack is different between training and inference, give it a read at https://arxiv.org/abs/2509.09747.

Great work by my two master students: Leen Daher and Zhaobo Wang.

#ICRA #machinelearning #science #sensors #robotics #SensorFusion