Anish Diwan

Anish Diwan is a Ph.D. student at the Intelligent Autonomous Systems Group of TU Darmstadt. His research focuses on

• Reinforcement Learning
• Imitation Learning
• Score-Based Generative Models

under partial observability constraints.

His work on Noise-conditioned Energy-based Annealed Rewards (NEAR) was published at ICLR 2025, introducing a novel framework for imitation learning from observation data.

Current research trends include

• Integration of score-based generative models with policy learning
• Spatial reward composition techniques
• Partial observability handling via energy functions

He utilizes Isaac Gym for GPU-accelerated robotics simulations and has contributed to open-source repositories implementing NEAR framework.