Sandeep Routray
Member of Technical Staff @ Skild AI
About
I'm a Master's student in Machine Learning at Carnegie Mellon University, working with Prof. Deepak Pathak on scaling robot learning using internet-scale videos without action labels. My research focuses on building general-purpose embodied agents that can operate in unstructured, real-world environments. I work on: • Leveraging Internet-scale video and language data for scalable, label-efficient robot learning • Using pretrained multimodal models for cross-task and cross-embodiment generalization • Employing video generation models to learn predictive world models for policy learning and planning • Learning actionable representations from complex, multimodal scenes (images, depth maps, 3D data) Previously, I worked at Samsung Research (CES 2024: Map View) and was a research fellow at the Vector Institute, focusing on self-supervised and object-centric learning. I completed my undergraduate degree in Electrical Engineering with a minor in Machine Learning from IIT Kanpur.
United States
San Mateo
Research
Robotics, Large Language Models (LLM), JAX, Generative AI, Fine Tuning, Software Industry, MySQL, SQL, Reinforcement Learning, PyTorch, CUDA, Deep Learning, Computer Vision, Machine Learning, Wireless Communications Systems, Python (Programming Language), C++, C (Programming Language), Natural Language Processing (NLP)
Experience
Member of Technical Staff
Skild AI
San Mateo, California, United States
• Building general-purpose robotics foundation model.
Technical Intern
Skild AI
Pittsburgh, Pennsylvania, United States
• Leveraging human videos to learn robot policies.
Graduate Research Assistant
Carnegie Mellon University
Pittsburgh, Pennsylvania, United States
• Research on adapting multimodal LLMs and video diffusion models to predict actions from internet-scale human and robot video datasets enabling few-shot task, environment and embodiment generalization
Machine Learning Engineer
Samsung Electronics
Seoul, South Korea
• Pioneered development of deep learning and vectorization stages of Map View engine to convert home layouts in wild (brochures, LiDAR, CAD, hand-drawn, etc.) to interactive 3D model. The engine was unveiled at CES 2024 and has been deployed across 400,000 homes globally. • Designed a ConvNeXt based model and trained using focal loss to identify rooms, walls, doors and junctions in layouts. Performed integer quantization and deployed in Android and iOS devices using TF Lite C API achieving 4x reduction in size and 3x increase in inference speed. • Worked on 3D reconstruction from single image by training a neural radiance field (NeRF) on multi-views generated from a diffusion model.
Research Fellow
Vector Institute
Toronto, Ontario, Canada
• Researched ways to leverage inter‑image relationships for object‑centric self‑supervised learning. Designed an image‑context aware score function to mine positives/negatives from a queue of images for contrastive loss. Obtained 1 % mIoU improvements over existing benchmarks. • Experimented ways to incorporate geometrical cues from 2D‑mesh decomposition of image and perform hierarchical grouping to discover object‑part relationships in a dense feature self‑supervised feature learning framework.
Intern
Samsung Electronics
Seoul, South Korea
• Implemented a reinforcement learning based resource scheduler for LTE system using Deep Deterministic Policy Gradient (DDPG) algorithm. • Devised two reward mechanisms to maximize throughput while maintaining QoS requirements of delay and fair allocation among users. • Achieved lower delay (upto 80% lower) and better scalability than the prevalent Proportional Fair scheduler without compromising data rates.
Research Intern
Indian Institute of Technology, Kanpur
Kanpur, Uttar Pradesh, India
Worked on developing height invariant object detection systems using unsupervised learning and domain adaptation.
Education
Carnegie Mellon University
Machine Learning
Indian Institute of Technology, Kanpur
Electrical Engineering
- Minor in Machine Learning - Awarded Academic Excellence Award 2017‑18, 2018‑19, 2019‑20 Equivalent to Dean’s List - Awarded Prof. Samares Kar Memorial Gold Medal for outstanding work in undergraduate project
Sandeep Routray's Contact Information
Phone
Find the Right Leads
Find Verified Contact Data
What LeadContact does well
Find verified emails, phone numbers, and decision-makers with 98% accuracy.
Find Leads
Find the right people by company, role, industry, location, and more.
925M+ professional profiles
Find Emails
Access verified email addresses for your target contacts.
657M+ emails
Find Phone Numbers
Get cross-validated phone data from multiple top sources.
239M+ phone numbers
More Accurate. Lower Cost.
Find contact data in 1 tool with 98% accuracy
LeadContact integrates leading enrichment tools to deliver more accurate contact data—without paying for each one.
Great conversations start with the right contact.
It’s time to find yours.