Cong Tai

Tsinghua–Berkeley Shenzhen Institute, University Town of Shenzhen

Papers

Total Citations

H-Index

About

Cong Tai is a rising researcher at the forefront of embodied AI and mobile robotics, with key contributions spanning robotic task planning, dynamic scene understanding, and visual relocalization. His most impactful work, "GRID: Scene-Graph-based Instruction-driven Robotic Task Planning" (2024, 19 citations), pioneers the use of scene graphs to bridge large language models (LLMs) with environmental context, enabling more precise instruction grounding for complex robotic tasks—a significant leap beyond raw-image-based approaches. To address a critical gap in robotics, Tai led the creation of the THUD (Tsinghua University Dynamic) dataset (2024, 11 citations), a large-scale indoor dataset designed specifically for evaluating robots in dynamic, real-world environments, moving beyond static benchmarks. Additionally, his FusedNet (2024, 4 citations) introduces a cross-attention mechanism that fuses global and local image features for end-to-end monocular relocalization, achieving robust performance in both static and dynamic scenes. With all major works published in 2024, Tai’s research is rapidly gaining traction, demonstrating a clear trajectory toward making robots more adaptive, context-aware, and reliable in unstructured settings.