Athrva Pandhare

I am a graduate student at University of Pennsylvania specializing in Robotics. I completed my undergraduate studies in Aerospace Engineering where I specialized in Computational Fluid Dynamics.

My interests lie within the areas of Autonomous Exploration, Robot navigation, Simultaneous Localization and Mapping (SLAM), Computer Vision and Machine Learning.

I am also a Student Researcher at the Kumar Robotics Lab at the University of Pennsylvania. I work in the field of Learning based Autonomous Exploration, and Applications of Deep Learning in Visual SLAM.

Feel free to check out my Resume and drop me an e-mail if you want to chat with me!

Email  /  Resume  /  Linkedin  /  Github

profile photo
Research
Point2Point: An Efficient Generative Architecture for Direct Point Cloud to Point Cloud translation
Athrva Pandhare
Submitted to IROS, 2023

/ Slides / Results Video

Inducing a Locality Consistent Ordering using Hilbert Space filling curves to facilitate direct learning on Unordered sets.

Multi-Scale RANSAC for Online Clutter removal from Indoor Point Clouds
Research Intern, Kumar Robotics (PERCH Lab, University of Pennsylvania), Summer 2022

Details /

Developed a Novel Algorithm for Clutter Removal from Egocentric Point Clouds.

Fast Path Planning with Local Minima Avoidance using Dynamic Position Penalty Artificial Potential Fields (DPPAPF)
Research Intern, Kumar Robotics (PERCH Lab, University of Pennsylvania), Summer 2022

Details /

Developed a Position Penalty APF for fast Path Planning in Indoor Environments

Generative Design of Airfoils for Vertical Axis Wind Turbine Applications using Evolutionary Algorithms
Athrva Pandhare, Arnab Paul, Senthilkumar Sundraraj
Bachelor's Thesis, 2020

Manuscript / Code

Developed Genetic Algorithms tailored for Airfoil Profile Optimization by maximizing Tangential force Coefficient

Comparative Study of Micro-nozzle characteristics under marginally rarefied and continuum treatments
Athrva Pandhare, Abhishek Puri, Senthilkumar Sundraraj
Bachelor's Independent Study, 2019

Manuscript /

Developed a Partial-Slip Based formuation for modelling transition from continuum to rarefied flow regimes.

Experience
Autonomous Exploration in Indoor Environments using Frontier based Exploration Strategies
Research Intern, Kumar Robotics (PERCH Lab, University of Pennsylvania), Summer 2022

Details /

Created Frontier Based Exploration algorithms for Autonomous Quadrotors in Indoor Environments

Learned Reinforcement Learning based direction primitives for Autonomous Exploration of Indoor Environments
Student Researcher, GRASP Lab, UPenn

Video / Details

Developed Direction Primitive based Reinforcement Learning models for Fast and Safe Autonomous Exploration

Monocular Depth Estimation using pix2pix GAN
Final Project, Machine Learning

Report / Code

An implementation of pix2pix GAN for Monocular Depth Estimation on KITTI outdoor scene data.

Image Semantic Segmentation using Attention U-Net with a study of encoded logits using a Combined Information Criterion
Final Project, Computer Vision and Computational Photography

Video / Report / Details

Attention UNet based segmentation with a Gradient Error minimization Loss and subsequent analysis of the Logits using Information criteria.

Adding Details to Images : Synthetic Outdoor Scence Generation
Independent Research, Generative Deep Learning

Video / Details

Outdoor Scene Generation from Semantic Information using pix2pix GANs for a small set of outdoor Environments

Region Based Unsupervised Image Segmentation and Simplification
Independent Research, Unsupervised Generative Deep Learning

Details /

Re-implementation of a Research paper on Unsupervised Image segmentation.

Semantic Segmentation of Outdoor Scenes using Adversarial training of a Efficient Net Based Generator and Patch-GAN based discriminator
Independent Research, Generative Deep Learning

Details /

Treating Semantic Segmentation as an Image Generation Task.

Generation of Raw Velodyne Point Cloud Projection from RGB Images using U-Net Architecture with ResNet based Encoder
Independent Research, Generative Deep Learning

Details /

Prediction Point Cloud Spherical Projection from Images in the KITTI raw dataset



Website adapted from Jon Barron .