Research Engineer
1X Technologies · California, USA
Worked with humanoid robots to make them capable of operating in real-world settings such as homes and warehouses.
Anant Rai
Hi, I’m Anant. I’m a founding research engineer at a stealth startup, working to make robots more capable and general-purpose so they can become part of daily life. Previously, I was among the first five members of the AI team at 1X Technologies; you can see some of that work here. Before that, I was a graduate research assistant at NYU’s CILVR Lab, advised by Lerrel Pinto and Soumith Chintala. Outside work, I take photographs.
Experience
Graduate Research Assistant
CILVR, New York University · New York, USA
Worked on imitation learning, representation learning, and generalizable reinforcement learning in collaboration with Hyundai, advised by Lerrel Pinto.
Research & Development Intern
Temasek Lab, Nanyang Technological University · Singapore
Led development of a meeting-room speech recognition Android application for automatic transcription, and built an image-captioning model using transformers.
Remote Research Assistant
CITEC Lab, Bielefeld University · Germany
Used conflict-based search and deep Q-learning on the Flatland environment, reaching third position in round one and sixth in round two.
Research and Publications
Science Robotics submission · 2023
On Bringing Robots Home
We developed an affordable tool for collecting robot demonstrations and used it to assemble the Homes of New York dataset: 13 hours of interaction across 22 homes. A self-supervised representation pretrained on this data improves robot policies in unseen home environments.
ICRA submission · 2023
Open X-Embodiment: Robotic Learning Datasets and RT-X Models
This work consolidates diverse robotic datasets to train generalist policies across robots, tasks, and environments, improving spatial understanding, generalization to unseen objects, and robustness over robot-specific baselines.
Best Student Paper · RSS 2023
Teach a Robot to FISH: Versatile Imitation from One Minute of Demonstrations
FISH is an imitation-learning approach that learns robust visual skills from less than a minute of human demonstrations. It works across robot morphologies and camera configurations while outperforming prior approaches.
IEEE ITSC · 2022
Risk Analysis using Trajectory Prediction in Indian Traffic
We combined detection, tracking, and an LSTM-based trajectory model to understand agent interactions in dense heterogeneous traffic, then introduced a weighted elliptical risk model that improved predictive risk analysis by 20% over the baseline.
Research project
RLMU: Reinforcement Learning with Self-supervised Human Motion Analysis
We made reinforcement learning from human videos more robust by adding a motion-understanding module that stabilizes cross-domain adaptation. On visual pushing, the approach improved performance by 15% over the baseline.
NeurIPS 2021 Flatland Challenge
Solving Multiagent Planning Problems using Deep Reinforcement Learning
We designed tree- and graph-based environment representations for intelligent cost calculation and rerouting. Combining conflict-based search with deep Q-learning reached third place in round one and sixth in round two.
ICAIR · 2020
Food Recommendation System using Neural Collaborative Filtering and Sentiment Analysis
We combined collaborative filtering, neural classification, and sentiment analysis to model customer food preferences using data collected from Swiggy and Zomato, then presented the results through a web application.