Karthik Narasimhan

Papers

Total Citations

H-Index

About

Karthik Narasimhan is a rising researcher at the intersection of natural language processing, reinforcement learning, and robotics, with a focus on building AI systems that are both capable and safe. His work addresses fundamental challenges in grounding language in physical environments and ensuring safe autonomous decision-making. In his highly cited paper on spatial reasoning, Narasimhan developed novel relation network architectures that enable robust and interpretable grounding of spatial references—a critical capability for tasks like autonomous navigation and robotic manipulation. This work, which has garnered significant attention, tackles the long-standing problem of learning multi-modal representations for spatial concepts without explicit supervision. Equally impactful is his pioneering research on safe reinforcement learning with natural language constraints. Recognizing that traditional safe RL methods require mathematical constraint specifications that demand domain expertise, Narasimhan proposed a groundbreaking framework that allows non-experts to specify safety requirements using natural language. This innovation, which has accumulated citations for its practical significance, dramatically lowers the barrier to deploying safe RL in real-world applications like autonomous driving and robotics. Through these contributions, Narasimhan is helping to bridge the gap between human communication and machine safety, making autonomous systems more accessible and trustworthy.