Hugo Caselles-Dupré

About

Hugo Caselles-Dupré is a researcher at the forefront of embodied AI, focusing on the intersection of continual reinforcement learning, robotics, and affordance perception. His major contributions address two critical challenges: enabling robots to learn multiple tasks sequentially without catastrophic forgetting, and grounding AI perception in actionable environmental understanding. In his seminal work "DisCoRL" (35 citations), he introduced a novel framework combining policy distillation with continual learning, allowing a single model to master diverse tasks and autonomously infer which policy to deploy—a breakthrough for real-world deployment. He further validated this approach in "Continual Reinforcement Learning deployed in Real-life using Policy Distillation and Sim2Real Transfer" (17 citations), demonstrating that robots can learn tasks sequentially without forgetting past skills, bridging the sim-to-real gap. More recently, Caselles-Dupré has explored affordance segmentation (2 citations), investigating whether standard object segmentation models can detect action possibilities in objects—a crucial step toward agents that understand not just *what* an object is, but *what can be done with it*. His work is foundational for building adaptive, lifelong-learning robots that operate robustly in dynamic human environments.

Research Focus

Key Achievements

Top Papers

1. 1
   DisCoRL: Continual Reinforcement Learning via Policy Distillation

   35 citations · 2019
2. 2
   Continual Reinforcement Learning deployed in Real-life using Policy Distillation and Sim2Real Transfer

   17 citations · 2019
3. 3
   Are standard Object Segmentation models sufficient for Learning Affordance Segmentation?

   2 citations · 2021

Key Collaborators

Contact & Links