Excavation Autonomy
Real-world excavation is one of the current application areas driving the lab's work in task-aware perception, robust visual sensing, and safe autonomous operation under field conditions.
Karthik Dantu
Karthik Dantu is an Associate Professor in Computer Science and Engineering at the University at Buffalo. He works on safe and efficient methods for perception, representation and planning for robot autonomy. His research is centered around two field deployments - culvert inspection and autonomous excavation. For his work, he has received the IEEE Region 1 Technological Innovation award (2022), NSF Early Career Award (2019), several nominations to best paper awards and the UB CSE senior researcher of the year in 2021. He is a senior member of the IEEE, member of ACM and AAAS. At UB, he is the Founding Director of the Center for Embodied Autonomy and Robotics, Co-Director of the MS program in Robotics and leads the Distributed RObotics and Networked Embedded Systems (DRONES Lab).
Projects
Culvert Inspection
Culvert inspection is the second major application theme currently highlighted in the lab: a good example of autonomy and perception designed for infrastructure settings rather than clean indoor demos.
Visual SLAM
- GPU-acceleration - Graphite [ICRA26], FastLoop [arXiv26], TurboMap [arXiv25], FastTrack [IROS25], JacobiGPU [ICRA24], Local Bundle Adjustment [ICRA23]
- Edge-Assisted SLAM - Edge-SLAM [MobiSys20, TECS22], PRE-SLAM [MASS22]
- Systems Challenges - A Quantitative Analysis of System Bottlenecks in Visual SLAM [HotMobile22], A Comprehensive Study of Systems Challenges in Visual Simultaneous Localization and Mapping Systems [TECS24]
Identification of Recyclable Plastics
- Thermal sorting - TIPS: Thermal Image based Plastics Sorting [MobiSys26]
- Quantifying recycled content - Detecting percentage of recycled plastic in plastic products [Nat Comm Eng26]
- Spectra - Mid-infrared spectroscopy and machine learning for postconsumer plastics recycling [ES:Adv23]
- REMADE - Improved Detection of Recyclable Plastics Using Multi Modal Sensing and Machine Learning [Scholar23]
Terrain Relative Navigation
- Tracking - Comparative Analysis of Feature Tracking Challenges in Unstructured Space Environments [TAES26]
- Landmark tracking - Domain Adaptation and Multi-view Attention for Learnable Landmark Tracking with Sparse Data [arXiv25]
- Terrain detection - You only crash once v2 [arXiv25]
- Feature recognition - MARs [ECCV24]
- Vision profiling - NASA SpaceCube profiling [AeroConf24]
- Hazard detection - You only crash once [arXiv23]
- Feature matching - Hypothesis Gating for TRN [SciTech22]
News
- 2026 Journal paper on scalable multi-UAV area coverage published in JCISE
- 2026 TAES paper on feature tracking challenges in unstructured space environments
- 2025 FastTrack accepted to IROS 2025
- 2024 MARs appeared at ECCV 2024
- 2024 Visual SLAM systems study published in ACM TECS