Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Publications
-
OptLayer - Practical Constrained Optimization for Deep Reinforcement Learning in the Real World
IEEE International Conference on Robotics and Automation (ICRA)
-
Hand-Object Contact Force Estimation From Markerless Visual Tracking
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI)
-
Multi-Contact Interaction Force Sensing from Whole-Body Motion Capture
IEEE Transactions on Industrial Informatics (TII)
-
Whole-Body Contact Force Sensing From Motion Capture
IEEE/SICE International Symposium on System Integration (SII 2016)
-
Towards Force Sensing from Vision: Observing Hand-Object Interactions to Infer Manipulation Forces
IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015)
-
Capturing and Reproducing Hand-Object Interactions Through Vision-Based Force Sensing
IEEE ICCV Workshop on Object Understanding for Interaction (OUI 2015)
Patents
-
Safe and fast exploration for reinforcement learning using constrained action manifolds
Filed US US20200065666A1
-
Imitation learning by action shaping with antagonist reinforcement learning
Filed US US20200034706A1
-
Sequential learning of constraints for hierarchical reinforcement learning
Filed US US20200034704A1
-
Action shaping from demonstration for fast reinforcement learning
Filed US US20200034705A1
Honors & Awards
-
SII 2016 Best Paper Award
-
-
Prix Mayoux-Dauriac Aéronautique (Master's Thesis Award)
-
Languages
-
French
Native or bilingual proficiency
-
English
Native or bilingual proficiency
-
German
Limited working proficiency
-
Japanese
Elementary proficiency
Explore more posts
-
Francois du Plessis
This is really interesting. The technology has been developing for a while now and it seems like it "popped" into the mainstream now. This is a diffusion large language model and it is - incredibly fast, therefore - takes must less time to process and therefore - uses much less energy. This has to be the direction that most of the LLM applications go. I also so the following Youtube review which gives some usage perspective: https://lnkd.in/dWWufjza
-
Yu Cao
Beyond Residual RL: Rethinking Model-Based Constraints for Truly Intelligent Robots After reading the RAMBO paper, I had a moment of déjà vu. Back in 2020, I experimented with combining MPC controllers with small RL networks—letting model-based methods handle the bulk, while RL “patched up” the messy bits. This hybrid approach was efficient and robust, but ultimately, it hit a ceiling: the RL agent’s creativity and adaptability were bottlenecked by rule-based constraints. Now, inspired by advances in LLM training, I’m rethinking this paradigm. What if we could fully distill our model-based “expert” into a neural policy using supervised fine-tuning (SFT)—and then unleash the power of RL to go beyond, just like RLHF did for language models? With a strong SFT initialization, RL could explore the vast landscape of robot behaviors, breaking free from rule-based limitations while retaining safety and efficiency. This two-step (SFT + RL) recipe is already redefining the future of robotics. It's not just patching up models—it’s about end-to-end learning, scalable adaptation, and new levels of autonomy. https://lnkd.in/eqjbPaXN #ReinforcementLearning #Robotics #ImitationLearning #SFT #AI #IntelligentControl #EndToEndLearning #PolicyDistillation #RLHF #AutonomousSystems
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content