RESEARCH
Making robots safe, robust and useful in our everyday spaces.
A multidisciplinary approach to AI.
Our main areas of expertise are computer vision (human understanding, human pose estimation, lifelong learning, 3D vision, representation learning…), natural language processing (speech, controlled generation, translation, multimodal neural search…), machine learning for robotics (reinforcement learning, manipulation, navigation, task learning…), machine learning for optimization (data driven optimization, deep learning for combinatorial optimization, stochastic programming…), human-robot interaction and software engineering to integrate AI/ML components with no code/low code. These different areas of expertise focus on one or several of the three AI for Robotics themes (action, interaction and vision).
DISCOVER
- CroCo: Self-supervised pretraining for 3D vision tasks by cross-view completion / SACReg: Scene-Agnostic Coordinate Regression for Visual Localization
- HUMANS: Learning visual models that can understand and predict human behaviour from images or videos.
- KAPTURE: A unified data format for visual localization, structure from motion and more.
- Proxy Virtual Worlds & Virtual KITTI 2
- DiPCAN: Distilling privileged information for crowd-aware navigation
- PoseScript: 3D human poses from natural language
- Continual adaptation of visual representations via domain randomization and meta-learning
- Concept generalization in visual representation learning
- StacMR: Scene-text aware cross-Modal Retrieval
- SMPLy benchmarking 3D human pose in-the-wild
- SLACK: Stable learning of augmentations with cold-start and KL regularization
- SPLADE: a sparse bi-encoder BERT-based model
- DISCO: a toolkit for DIStributional COntrol of generative language models
- Lifelong Representation Learning (Research Chair – MIAI Institute)
- Online Adaptation for Semantic Image Segmentation (OASIS)
- Improving the generalization of supervised models(t-ReX)
- ARTEMIS: Attention-based retrieval with text-explicit matching and implicit similarity
- Deep Image Retrieval
- Hard Negative Mixing for Contrastive Learning (moCHI)
- Efficient multilingual machine translation
- Human-Robot Interaction on Medium
