Faraz Torabi

Papers

Total Citations

H-Index

About

Faraz Torabi is a researcher whose work sits at the intersection of imitation learning and reinforcement learning, with a particular focus on enabling agents to learn complex behaviors without direct access to expert actions. His primary research areas include imitation from observation, inverse dynamics modeling, and sample-efficient adversarial learning. Torabi’s major contribution is the development of RIDM (Reinforced Inverse Dynamics Modeling), a novel framework that allows an agent to learn from a single observed demonstration by augmenting reinforcement learning with imitation learning—significantly reducing the assumptions typically required for such integration. His work on adversarial imitation from observation has also advanced the field by improving sample efficiency, achieving notable performance gains over traditional methods. With his most-cited paper accumulating 11 citations, Torabi’s research has helped bridge the gap between demonstration-based learning and autonomous policy acquisition. His 2019 survey on recent advances in imitation learning from observation remains a valuable resource for researchers entering the field, offering a comprehensive overview of state-only imitation techniques. Torabi’s contributions are particularly impactful for robotics and autonomous systems, where learning from human demonstration without action labels is a critical challenge.