Embedded Files
Autonomy
Autonomy
This is my primary research interests align with creating robust and safe autonomous systems of reliable and trustworthy performance.
Engineering of the Field
Engineering of the Field
I also have worked with research for specific use in practical scenarios in the engineering field. This was boosted my experience in Autonomy research also studies in Mechanical Engineering in college.
Autonomy & Control
Safe Learning Algorithm with fast adaptation for non-stationary environments.
Safe Learning Algorithm with fast adaptation for non-stationary environments.
Efficient Data-driven Offline Learning by inducing Additional Conservatism via Sparsity
Efficient Data-driven Offline Learning by inducing Additional Conservatism via Sparsity
- Abstract -
To facilitate the practical use of AI in safety-critical domains like autonomous driving and robotic manipulation, I studied the unique challenges of ensuring safety in non-stationary environments by solving constrained problems through the lens of the meta-learning approach.
- Contribution -
Suggest the efficient framework for model-free constrained meta-RL
The theoretical contribution that enables performance guarantees in constrained meta-learning settings
Empirical studies that justify our study with outperforming performance over others
Preprint: https://arxiv.org/pdf/2312.10230.pdf [Accepted at AAAI-24]
- Abstract -
In offline learning settings, optimization without safety conservatism leads to safety violations, requiring the identification of cost upper bounds or the introduction of penalty terms. Previously, this was done by data perturbation with a bi-level structure which may introduce instability in high-dimensional tasks. In this work, we identify potentially hazardous regions based on the sparsity of given data.
- Contribution -
Suggest a simple alternative way to introduce safety-conservatism
Provide empirical studies showing simplicity and promising performance.
Preprint: Preparing
- Abstract -
In offline learning settings, optimization without safety conservatism leads to safety violations, requiring the identification of cost upper bounds or the introduction of penalty terms. Previously, this was done by data perturbation with a bi-level structure which may introduce instability in high-dimensional tasks. In this work, we identify potentially hazardous regions based on the sparsity of given data.
- Contribution -
Suggest a simple alternative way to introduce safety-conservatism
Provide empirical studies showing simplicity and promising performance.
Preprint: Preparing
PowerPoint: here
Description
Out-of-distributional Adaptation in Offline Learning via Causal Normalizing Flow
Out-of-distributional Adaptation in Offline Learning via Causal Normalizing Flow
Coming Soon!! (Thinking as Offline, Robust, Safe, Causal RL)
Coming Soon!! (Thinking as Offline, Robust, Safe, Causal RL)
Description
Engineering Application
Development of Parameter Dependent GAN for Multi-Phase Flow Prediction
Development of Parameter Dependent GAN for Multi-Phase Flow Prediction
Interpolation Schemes of Richardson Extrapolation for High-accuracy Solution of Wave Equations (Poisson & Hemlholtz)
Interpolation Schemes of Richardson Extrapolation for High-accuracy Solution of Wave Equations (Poisson & Hemlholtz)
- Abstract -
Poor air-fuel mixing causes deterioration in combustion engines’ performance and emissions. While designers rely on computational fluid dynamics (CFD) modeling to understand the air-fuel mixing process, there are recognized shortcomings in current CFD predictions due to a lack of phenomenological models. To overcome this, we employed ML to efficiently predict the gasoline fuel sprays under realistic engine conditions.
- Contribution -
Show the potential of using the ML algorithm for spray prediction as an alternative to CFD.
Suggest the general framework and conditions to create realistic engine experimental conditions in virtual software for ML integration.
Preprint: google drive [Under development]
- Abstract -
Employing wave equations as a Partial Differential Equation (PDE) is common in the oil exploration industry, where waves are utilized to estimate the location and quantity of oil. However, the disturbances and instability in estimating wave oscillations pose challenges in achieving accurate estimations. Therefore, we propose various Richardson Extrapolation interpolation schemes to offer robust and precise solutions for practical applications in the industry.
- Contribution -
Provide an overview of Richardson Extrapolation in wave estimation
Analyze key aspects of using interpolation schemes for higher-order solutions that achieve super-convergence over others.
Preprint: Preparing
Page updated
Google Sites
Report abuse