Angel Daruna

☕️

Angel Andres Daruna

Advanced Computer Scientist II

About Me

Angel Daruna is a computer scientist with the SUN lab of the SRI Center for Vision Technologies. His research interests include machine learning, geospatial intelligence, robotics, and knowledge graphs.

Download Extended CV

Interests

Machine Learning
Geospatial Intelligence
Knowledge Graphs & Robotics

Education

PhD Robotics
Georgia Insitute of Technology
MSc Computer Science
Georgia Insitute of Technology
BSc Computer Engineering
Georgia Institute of Technology

Featured Publications

Self-Supervised Learning

Enabling Scalable Mineral Exploration: Self-Supervision and Explainability

This work introduces a unified framework for Mineral Prospectivity Mapping that combines self-supervised learning, explainable AI, and uncertainty modeling to overcome challenges in deep learning caused by limited labeled data and interpretability. The approach improves prediction robustness and provides geologically meaningful insights, demonstrated through evaluations on Zn–Pb deposits in North America and Australia.

Jan 1, 2024

Explainable AI

Explainable knowledge graph embedding: Inference reconciliation for knowledge inferences supporting robot actions

This work proposes an inference reconciliation framework that uses decision tree-based local approximations and natural language explanations to interpret black-box knowledge graph embeddings driving robot behavior. Evaluations show that the approach helps non-experts understand and correct erroneous robot actions caused by flawed internal beliefs.

Jan 1, 2022

Continual Learning

Continual Learning of Knowledge Graph Embeddings

This work addresses the challenge of incorporating new semantic knowledge into distributed representations by applying continual learning methods to incremental knowledge graph embedding. It offers an experimental evaluation highlighting trade-offs to help practitioners choose appropriate methods for semantics-driven robotic applications.

Jan 1, 2021

Robotics & Automation

Taking recoveries to task: Recovery-driven development for recipe-based robot tasks

This work introduces Recovery-Driven Development (RDD), an iterative task scripting approach that simplifies the development of robust robot systems by separating task and recovery logic and enabling situated testing. The effectiveness of RDD is demonstrated through its successful application in a challenge-winning mobile manipulator at the IEEE 2019 FetchIt! Challenge.

Jan 1, 2019

Recent Publications

Han-Pang Chiu, Yi Yao, Angel Daruna, Yunye Gong, Abhinav Rajvanshi, Giedrius Burachas (2024). UNCERTAINTY ESTIMATION FOR NEURAL NETWORKS USING GRAPHICAL REPRESENTATION.

Cite

Angel Daruna, Vasily Zadorozhnyy, Georgina Lukoczki, Han-Pang Chiu (2024). Enabling Scalable Mineral Exploration: Self-Supervision and Explainability. 2024 IEEE International Conference on Big Data (BigData).

Cite

Angel Daruna, Vasily Zadorozhnyy, Georgina Lukoczki, Han-Pang Chiu (2024). GFM4MPM: Towards Geospatial Foundation Models for Mineral Prospectivity Mapping. Proceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems..

Cite

Weiyu Liu, Angel Daruna, Maithili Patel, Kartik Ramachandruni, Sonia Chernova (2023). A survey of semantic reasoning frameworks for robotic systems. Robotics and Autonomous Systems.

Cite

Weiyu Liu, Dhruva Bansal, Angel Daruna, Sonia Chernova (2023). Learning instance-level n-ary semantic knowledge at scale for robots operating in everyday environments. Autonomous Robots.

Cite

See all publications

🎉 Coauthored Proposal and Participated in DARPA CriticalMAAS

Feb 1, 2024

Served as technical lead to design machine learning software for mineral potential mapping exploiting multi-modal fusion

Feb 1, 2024