Eric Kolve

Papers

Total Citations

1,677

H-Index

About

Eric Kolve is a leading researcher in Embodied AI, whose work bridges computer vision, deep reinforcement learning, and robotics to create intelligent agents that can perceive, navigate, and interact with complex 3D environments. He is best known for pioneering **target-driven visual navigation**, a paradigm that enables agents to generalize to new goals without retraining—a critical advance over earlier reinforcement learning approaches that were data-inefficient and brittle. His seminal 2017 paper, *"Target-driven visual navigation in indoor scenes using deep reinforcement learning,"* has garnered over **1,500 citations**, establishing a foundational framework for goal-conditioned policies in simulated indoor spaces. Building on this, Kolve contributed to **ManipulaTHOR**, a framework that extends Embodied AI from navigation to active object manipulation, enabling agents to physically interact with their surroundings. His work has been instrumental in shaping the AI2-THOR simulation platform, a widely adopted benchmark for embodied research. By tackling the twin challenges of generalization and data efficiency, Kolve has helped move AI agents from passive observers to active participants in virtual worlds, laying the groundwork for real-world robotic applications.