Akshara Rai

Akshara Rai is a robotics researcher whose published work applies machine learning to robot control and embodied interaction. Her research includes reactive movement and simulation-based bipedal control, and she completed a Carnegie Mellon robotics PhD in 2018. Her publications span visual demonstrations, collaborative home simulators, household robot coordination, embodied planning, and multi-fingered manipulation. As of October 2026, she is co-founder and Chief Scientist at Versor, where she tackles robotics challenges through real2sim2real.

Now at: Versor
Works: 51
Citations: 973
h-index: 20
Badges: Founder
Bridge

Publication numbers from OpenAlex, as of 18 October 2026.

Timeline

2016

Learning reactive motion

In 2016, she coauthors a method that uses a neural network to learn reactive feedback for movement plans. The team tests the method on obstacle avoidance with a robot arm.
2018

Bipedal control and a PhD

In 2018, she coauthors a method that uses simulation to improve Bayesian optimization for bipedal robot controllers. She completes her PhD at Carnegie Mellon’s Robotics Institute in December 2018 with a thesis on learning from simulation.
2020

Learning from visual demonstrations

In October 2020, she coauthors a model-based inverse reinforcement-learning framework that learns cost functions from visual human demonstrations. The team uses those costs to reproduce demonstrated behavior with visual model-predictive control and tests the method on two manipulation tasks.
2023

Building Habitat 3.0

In October 2023, she coauthors Habitat 3.0, a simulator for collaborative human-robot tasks in homes. It supports human-in-the-loop evaluation and studies social navigation and rearrangement.
2023

Social robot coordination

In 2023, she coauthors a study of two robots coordinating on household tasks through onboard, egocentric observations. Its Behavior Diversity Play method helps agents generalize to new partners in unseen environments.
October 2024

The PARTNR benchmark

In October 2024, she coauthors PARTNR, a benchmark for human-robot collaboration on household activities. The benchmark contains 100,000 natural-language tasks across 60 houses and 5,819 unique objects.
August 2025–November 2025

Sim2Real research and AINA

In August 2025, she presents “Sim2Real Learning for Home Robots” as a Research Scientist at Meta. In November 2025, she coauthors AINA, a framework for learning multi-fingered robot policies from everyday human tasks using Aria Gen 2 glasses.
As of October 2026

Co-founding Versor

As of October 2026, she is co-founder and Chief Scientist at Versor, where she tackles robotics challenges through real2sim2real.