Context-Aware GAN-based Image Retrieval for Relocalization of Autonomous Robots, Ruphan Swaminathan and Pradyot Korupolu.
IEEE International Conference on Intelligent Robots and Systems (IROS), 2024.
Effective localization is crucial for the reliable operation of autonomous robots. This paper introduces ConLocGAN, a novel context-aware GAN, addressing challenges in Lidar-based localization. ConLocGAN extracts robust localization-specific global descriptors for coarse pose estimation, which acts as a precursor for Lidar-based pose refinement.

Computational design of ultra-robust strain sensors for soft robot perception and autonomy,
Haitao Yang, Shuo Ding, Jiahao Wang, Shuo Sun, Ruphan Swaminathan, Serene Wen Ling Ng, Xinglong Pan and Ghim Wei Ho.
Nature Communications, 2024.
Compliant strain sensors are crucial for soft robots’ perception and autonomy. However, their deformable bodies and dynamic actuation pose challenges in predictive sensor manufacturing and long-term robustness. This necessitates accurate sensor modelling and well-controlled sensor structural changes under strain. This work presents a computational sensor design featuring a programmed crack array within micro-crumples strategy.

MobileDeRainGAN: An Efficient Semi-Supervised Approach to Single Image Rain Removal for Task-Driven Applications,
Ruphan Swaminathan and Pradyot Korupolu.
IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPR), 2023.
Reliable operation of autonomous robots demand robust performance in adverse weather. This work addresses the following key limitations in the literature. First, the lack of incorporating real-world datasets during training results in a domain gap. Second, using computationally expensive forward passes impedes real-time performance. Third, the negligence in evaluating application-oriented metrics.

Unsupervised Intelligent Pose Estimation of Origami-Inspired Deployable Robots,
Rohit Lal, Ruphan Swaminathan, C.A.O.Sifan, Sishen Yuan, Lalith, Qui Liang and Hongliang Ren.
Chapter in: Deployable Multimodal Machine Intelligence, Lecture Notes in Bioengineering, Springer, 2023.
Origami is the art of folding paper into different shapes and structures, implying wide interdisciplinary usage. Vision-based structural tracking can help determine the suitable control and intelligent actions required based on the pose of the origami robot. This chapter deals with various unsupervised learning methods for multi-DOF pose estimation using a single camera as an attempt to do general-purpose origami tracking.

Multiphysics Simulation of Magnetically Actuated Robotic Origami Worms,
Ruphan Swaminathan, Catherine Jiayi Cai, Sishen Yuan and Hongliang Ren.
IEEE Robotics and Automation Letters (RAL) and International Conference on Robotics and Automation (ICRA), 2021.
Multiphysics simulation of magnetically actuated robots promises a range of applications such as synthetic data generation, design parameter optimization, predicting the robot’s performance, and implementing various control algorithms, but has rarely been explored. This work presents a realistic multiphysics simulation of magnetically actuated origami robots with focus on real-time interactions between the origami and magnets. Due to the interaction between multiple magnets, the complex motion dynamics of a worm-like robot are generated and analyzed.

ScoopNet: A 6DOF Pose Estimation Pipeline for Origami-inspired Worm Robots,
Rohit Lal, Ruphan Swaminathan, Lalithkumar Seenivasan, Liang Qiu and Hongliang Ren.
IEEE International Conference on Development and Learning (ICDL), 2021.
Origami-inspired soft and flexible robots have drawn immense attention in recent years for their wide range of medicine and engineering applications. While the ability of shape morphing presents a significant advantage, the shape-invariant pose estimation techniques are still under-explored. Pose estimation and tracking are vital to study, control and automate the locomotion of origami robots. This paper proposes ScoopNet that performs semantic segmentation and 6DOF pose estimation of origami-inspired worm robots.

A CNN-LSTM-based fault classifier and locator for underground cables,
Ruphan Swaminathan, Sanhita Mishra, Aurobinda Routray and Sarat Chandra Swain.
Springer Neural Computing and Applications, 2021.
Faults in underground cable systems are challenging to locate for maintenance and replacement due to their non-transient nature, which complicates accurate fault location, especially on long lines. To address this issue, this work proposes a deep learning algorithm that identifies the inception time, type, and location of faults for effective maintenance. Our experiments demonstrate that the network is robust across varying system parameters and, on average, can pinpoint faults within approximately 20 meters.