Carlos Florensa

About

Carlos Florensa is a leading researcher in reinforcement learning (RL) and robotics, whose work focuses on enabling agents to learn complex, goal-oriented tasks with greater efficiency and adaptability. His most impactful contribution, the seminal paper "Reverse Curriculum Generation for Reinforcement Learning" (2017), has garnered over 140 citations. This work introduced a powerful method for training agents in sparse-reward environments by automatically generating a curriculum of increasingly difficult starting states, a breakthrough that has become foundational for tackling manipulation tasks like assembly and lock insertion. Florensa further advanced the field by pioneering self-supervised learning of image embeddings for continuous control (2019), enabling robots to operate directly from raw visual inputs without hand-crafted reward functions. His research also addresses the critical challenge of sample efficiency and robustness, as seen in his work on guided uncertainty-aware policy optimization (2020) and adaptive variance for changing environments (2019). By blending model-based strategies with learning, Florensa's contributions are paving the way for robots that can learn, adapt, and generalize in the real world.

Research Focus

Key Achievements

Top Papers

1. 1
   Reverse Curriculum Generation for Reinforcement Learning

   140 citations · 2017
2. 2
   Self-supervised Learning of Image Embedding for Continuous Control

   30 citations · 2019
3. 3
   Guided Uncertainty-Aware Policy Optimization: Combining Learning and Model-Based Strategies for Sample-Efficient Policy Learning

   5 citations · 2020
4. 4
   Adaptive Variance for Changing Sparse-Reward Environments

   2 citations · 2019

Key Collaborators

Contact & Links