Nakul Agarwal

Papers

Total Citations

H-Index

About

Nakul Agarwal is a researcher advancing the frontiers of embodied AI and autonomous systems, with a core focus on motion prediction, spatial action localization, and interpretable machine learning. His work addresses critical challenges in human-robot collaboration and autonomous navigation by enabling machines to anticipate and reason about future human actions and agent behaviors. Agarwal introduced the novel task of spatial action localization in future frames, proposing "AdamsFormer" (2023, 7 citations), a transformer-based architecture that predicts where actions will occur in upcoming video frames—a vital capability for safe, proactive robot interaction. He further contributed to interpretable AI with his work on "Disentangled Neural Relational Inference" (2023, 4 citations), which improves motion prediction for dynamic agents by disentangling latent relational factors, offering transparency in how autonomous systems model interactions. Though early in his career, Agarwal’s research is notable for bridging predictive accuracy with model explainability, a crucial step toward trustworthy autonomous systems. His work is increasingly cited in the robotics and computer vision communities, positioning him as an emerging voice in next-generation motion planning and human-aware AI.