Papers
13
Total Citations
419
H-Index
8
About
Nathan Hughes is a robotics researcher whose work centers on spatial perception, hierarchical environment representations, and autonomous navigation. He is best known for developing **Hydra**, a real-time system for constructing and optimizing 3D scene graphs—layered graph structures that encode environments at multiple levels of abstraction, from raw geometry to high-level semantics like objects and rooms. First introduced in 2022 and accumulating over 170 citations, Hydra has become a landmark contribution to robot mapping and scene understanding. Hughes extended this framework through **Hydra-Multi**, enabling collaborative scene graph construction across multi-robot teams, and through **Clio**, which integrates open-set semantic models like CLIP and SegmentAnything to support task-driven, class-agnostic mapping. His 2024 survey on the foundations of spatial perception for robotics further consolidates the field's progress and directions. Beyond mapping, Hughes has explored how 3D scene graphs can power navigation policies via graph neural networks, demonstrating that rich hierarchical representations meaningfully improve robot decision-making. His work on **Kimera** and **Kimera2** bridges SLAM and semantic scene understanding in dynamic real-world environments. With nearly 400 total citations across his most impactful publications, Hughes has established himself as a leading voice in building richer, more actionable representations of the world for autonomous robots.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4<i>Clio:</i> Real-Time Task-Driven Open-Set 3D Scene Graphs35 citations · 2024
- 5
- 6
- 7
- 8Kimera: From SLAM to spatial perception with 3D dynamic scene graphs12 citations · 2021
- 9Kimera2: Robust and Accurate Metric-Semantic SLAM in the Real World7 citations · 2024
- 10Dynamic Grasping with a "Soft" Drone: From Theory to Practice2 citations · 2021