Abhishek Jha

I'm computer science graduate student at New York University, Courant. I completed my bachelor's from Delhi Technological University in New Delhi. My interests span computer vision, robotics, bayesian machine learning, and reinforcement learning, with a focus on dexterous manipulation and building reliable, interpretable decision-making systems across modalities.

Email  /  CV  /  Scholar  /  Github

Research

My research centers on building embodied agents that can perceive, reason, and act reliably in complex physical environments. I work at the intersection of dexterous manipulation, multi-agent control, and robot learning, studying how agents can plan verifiable actions, satisfy safety constraints, and scale gracefully across crowded, partially observed settings. A central theme running through my work is closing the loop between perception and control, where raw visual and tactile observations are transformed into temporally consistent, physically grounded behavior that generalizes across tasks and morphologies. I am currently working on dexterous manipulation and robot learning, studying how agents acquire and execute in-hand skills through reinforcement learning by fusing visual and tactile observations with 3D scene understanding and object pose estimation. Further working on developing VLA models as planners that ground open-vocabulary instructions into physically executable behavior for visual navigation where I explore how agents can interpret semantic and language-based goals, build spatial awareness from visual context, and move through unstructured environments without relying on pre-built maps. I also worked on multi-agent coordination, studying how large populations of agents can navigate shared spaces safely, resolve deadlocks, satisfy input constraints, and scale without centralized communication. Alongside this studying the developments that could be done in theoretical side of reinforcement learning, particularly in what makes learned policies sample efficient, robust under distribution shift, and transferable from simulation to real hardware. My broader goal is to develop agents that do not merely execute fixed skills but reason about their actions, adapt online to novel constraints, and operate safely in shared and unstructured environments.

A survey of Social Mini-Games in multi-robot navigation, proposing a unified taxonomy and evaluation framework to classify existing methods and guide future research.

A single decentralized control algorithm using neural ICBFs and gradient optimization that ensures safe, input-constrained, deadlock-free control of 1000+ agents in cluttered environments.

A semi-supervised segmentation model (PV-S3) that detects defects in photovoltaic EL images using only 20% labeled data, outperforming state-of-the-art supervised methods and reducing annotation costs by 80%.

A fine tuned transfer learning model using the SimCLR and SwAV models that predicts autism from resting-state fMRI scans, showcasing the potential of contrastive and non-contrastive models for robust neuroimaging analysis.

Introduced Strategic Pseudo-Goal Perturbation (SPGP), that resolves deadlocks in multi-agent navigation by guiding agents through strategic pseudo-goals, enhancing safety and efficiency in complex scenarios.

A transfer learning model using the Xception ConvNet predicts autism from resting-state fMRI scans, demonstrating the feasibility of early diagnosis through deep learning on brain imaging data.

An integrated ANN and Genetic Algorithm model predicts and optimizes Material Removal Rate and surface roughness in Al 6061 turning operations, enhancing machining precision through simulation-based methods.

Projects

A Benchmark and Survey of Deadlock Resolution in Multi-Robot Navigation in Social Mini-Games

Autonomous navigation and trajectory planning of a robot using Robot Operating System (ROS). A maze is created in gazebo for the robot to determine the best possible trajectory with collison avoidance. Probablistic localization method is used for navigation. Adaptive Monte Carlo Localization(AMCL) node and slam_gmapping package is used for localization of robot and mapping of robot. Rviz interface is used for the simulation of robot and creating the cost map for the travel of robot.

Obstacle avoidance implemented in Robot operating system for Unmanned Aerial Vehicle (UAV) using LiDAR scan data. Hector quadrotor package is used for spawning the drone in the gazebo environment. Readings from SONAR sensor with python script is used for flying the drone to a certain height. LiDAR readings are used to detect obstacles in the surroundings.

Path planning of point robot for finding shortest path between start and goal position. Implemented Informed and Uninformed search algorithms for the path planning problem. A random 2D enivronment is created with obstacles for evealuation of searching algorithms. A* and Dijkstras algorithm is implemented for obtaining the shortest path. Investigated the perofrmance by implementing other search based algorithms such as Best first, Depth First and Breadth First for finding the shortest distance. Given below are the images of search done by the various algorithms.

This website is created upon Jon Barron's template.