Search NASA⌕ Search

Engineering topics

Loc Tran

Publications and source records attributed to Loc Tran.

9 records
All sourcesNASA NTRSDOE OSTI
NASA NTRS2020

Search and Rescue under the Forest Canopy using Multiple UAS

We consider the problem of multi-robot search and rescue under the forest canopy. Forest is a particularly challenging environment for collaborative mapping and exploration, mainly due to the existence of severe perceptual aliasing, which hinders reliable mutual localization and map fusion. Our proposed system features unmanned aerial vehicles (UAVs) with onboard sensing and autonomy. Each UAV runs a lightweight filtering algorithm for local state estimation, and a dynamic-aware frontier selection algorithm for fast exploration. The essential and computationally intensive task of collaborative simultaneous localization and mapping (CSLAM) is performed at a central ground station. To handle perceptual aliasing, we make use of stable landmarks extracted from trees, which significantly improve precision and recall during place recognition. Furthermore, to recover from incorrect pairwise data associations during loop closure, we propose a novel procedure for global data association based on recently developed techniques on cycle consistent multiway matching. Our algorithm returns a global data association that is guaranteed to be cycle consistent, and is shown to significantly improve precision compared to the input pairwise associations. The overall multi-UAV system is extensively validated during real-world collaborative exploration missions in a forest at NASA Langley Research Center.

Multi-robot systems↗
NASA NTRSDate not supplied

Enhancing Neural Network Decision-Making with Variational Autoencoders

Machine intelligence has been used to tackle increasingly complex problems and deep learning solutions are at the forefront of tackling these problems. In general, these architectures have a great number of parameters that are methodically updated in training. The vast number and complexity of deep neural networks makes it very difficult to decipher the inner workings of the neurons and layers that make up the network. This paper posits that trustworthiness and trust in autonomous systems are increased through eXplainable Artificial Intelligence (XAI) and presents a method that enhances the explainability and understanding of a neural network decision. We leverage variational autoencoders to produce human interpretable features from complex data sets. We show that the explainable features can then be used for machine learning applications. This explainability encourages people to be more inclined to justifiably trust machine decision-making.

Loc Tran↗
NASA NTRSDate not supplied

Enhancing Neural Network Explainability with Variational Autoencoders

Machine intelligence has been used to tackle increasingly complex problems and deep learning solutions are at the forefront of tackling these problems. In general, these architectures have a great number of parameters that are methodically updated in training. The vast number and complexity of deep neural networks makes it very difficult to decipher the inner workings of the neurons and layers that make up the network. This paper posits that trustworthiness and trust in autonomous systems are increased through eXplainable Artificial Intelligence (XAI) and presents a method that enhances the explainability and understanding of a neural network decision. We leverage variational autoencoders to produce human interpretable features from complex data sets. We show that the explainable features can then be used for machine learning applications. Explainability inspires trust in autonomous systems that use deep learning, which is necessary for safety critical systems.

Loc Tran↗
NASA NTRSDate not supplied

Development of the Model Deformation System at the National Transonic Facility

This paper describes the model deformation system currently being developed for the National Transonic Facility at NASA Langley Research Center. The photogrammetry system uses up to eight network cameras that are time synced with an external hardware trigger. Camera extrinsics are calculated per frame with a bundle adjustment process with coded targets painted to the floor and walls of the test section. We describe our algorithm pipeline including the coded target system for target correspondence between camera views. We also present our future plans for modernizing the system.

Timothy Fahringer↗
NASA NTRSDate not supplied

Development of the model deformation system at the National Transonic Facility

This paper describes the model deformation system currently being developed for the National Transonic Facility at NASA Langley Research Center. The photogrammetry system uses up to eight network cameras that are time synced with an external hardware trigger. Camera extrinsics are calculated per frame with a bundle adjustment process with coded targets painted to the floor and walls of the test section. We describe our algorithm pipeline including the coded target system for target correspondence between camera views. We also present our future plans for modernizing the system.

Timothy Fahringer↗