Abrar Anwar

Hi! I'm an Ph.D student at the University of Southern California, where I work with Jesse Thomason on language grounding and human-robot interaction.

Previously, I was a student at UT Austin, where I worked on human-robot interaction with the Building-Wide Intelligence Project with Dr. Justin Hart, where I focused on how robots can leverage human behavior and how humans can understand robot behavior through cues such as eye gaze. Additionally, I finished an honors thesis on deep reinforcement learning for mesh sequence refinement supervised by Dr. Chandrajit Bajaj.

I previously interned at at Sandia National Laboratories with the Neural Exploration and Research Lab. In 2019, I worked with Dr. James Aimone on brain-inspired navigation for a hypersonic glide vehicle. In 2020 and 2021, I have been working with Dr. Craig Vineyard to evolve sparse, nosie resilient, spiking neural network topologies.

Email  /  GitHub  /  LinkedIn  /  Twitter  /  Scholar
Books  /  CV

profile photo



I'm interested in robotics, machine learning, neural networks, and computer vision.

project image

Deep Reinforcement Learning for Optimal Refinement of Cross-Sectional Mesh Sequence Finite Elements

Abrar Anwar
Honors Thesis, 2021

Advised by Dr. Chandrajit Bajaj . Developed the first deep reinforcement learning framework for mesh refinement, and refined “good” quality surface reconstructions of cross-sectional contours using soft-actor critic with initial simulations

project image

Watch Where You're Going! Gaze and Head Orientation as Predictors for Social Robot Navigation

Blake Holman, Abrar Anwar, Akash Singh, Mauricio Tec, Justin Hart, Peter Stone
International Conference on Robotics and Automation, 2021

Leverage virtual reality to collect gaze and position data to create a predictive model and a mixed effects model to show gaze orientation precedes other features

project image

Do you see what I see? Gaze understanding in people, 3D-rendered robot heads, and virtual reality

Akash Singh, Abrar Anwar, Akash Singh, Justin Hart
UT Undergraduate Research Forum, 2021
poster /

(Won best CS poster) We hypothesize that many virtual robot heads lose accuracy due to inaccuracies in the way that they compute the robot’s gaze direction, rather than the transfer from 3D to 2D, so we design an experiment to compare study participants’ interpretations of gaze direction on various platforms

project image

Evolving Spiking Circuit Motifs using Weight Agnostic Neural Networks.

Abrar Anwar, Craig Vineyard, William Severa, Srideep Musuvathy, Suma Cardwell
AAAI-21 Undergraduate Consortium , 2021
Computer Science Research Institute Summer Proceedings, 2020
International Conference on Neuromorphic Systems
, 2020
poster / tech report / paper /

An evolutionary, weight agnostic method is used to generate spiking neural networks used for classification, control, and various other tasks

project image

Using Human-Inspired Signals to Disambiguate Navigational Intentions

Abrar Anwar, Blake Holman, Connor Sheehan, Jeffery Huang
UT Undergraduate Research Forum, 2020
poster /

Specific navigational cues are used to study how users interact with various signals on our BWIBots in a hallway scenario.

project image

BrainSLAM: Robust autonomous navigation in sensor-deprived contexts

Felix Wang, James B. Aimone, Abrar Anwar, Srideep Musuvathy.
Sandia National Labs Technical Report, 2019
tech report /

Brain-inspired approaches to navigation and localization are explored in a noisy, data-sparse environment for a hypersonic glide vehicle. Rotation invariant feature representation methods are explored to increase accuracy and reduce map storage.

project image

Bounding Box SLAM: A Fast, Selective SLAM

Abrar Anwar, Blake Holman, Misha Shaposhnikov
UT Undergraduate Research Forum, 2019
code / poster /

Semantic information is combined with ORB-SLAM in order to reduce drift and localization error in dynamic environments.

Other Projects

These include coursework, side projects and unpublished research work.

project image

Calibrated Feedback for Language-Guided Reinforcement Learning

Clara Cannon, Abrar Anwar
Class Research Project - Advanced ML
paper /

Increased RL agent performance on reward-sparse Montezuma’s revenge by combining research on neural net uncertainty calibration and language feedback to develop a model-based interactive RL algorithm

project image

Detecting Muscle Cocontraction Through Sliding Window Gaussian Processes

Abrar Anwar
Class Project - Grad Machine Learning
paper / code /

Implemented a Gaussian process (GP) from scratch in order to detect muscle cocontraction in the hyperparameters in a set of sliding window GPs

project image

DeepHHD: Learning Helmholtz-Hodge Decomposition to Predict Optical Flow

Abrar Anwar
Class Project - Geometric Foundations of Data Science
paper / code /

By treating an optical flow estimate as a vector field, we can use a deep neural network to estimate the Helmholtz-Hodge Decomposition, whose sum is the optical flow itself.

Design and source code from Jon Barron's website