Hi! I'm currently in Carnegie Mellon's MRSD program, where I've gone from someone who worked in classical robotics to someone who now "teaches" robots how to do things.
This past summer, I interned at Bosch Center for AI, where I looked into failure modes in VLAs and hallucination in world models.
My work sits at the intersection of research and deployment, which allows me to reason through the "why" behind a system's behavior while staying grounded in what it actually takes to build and ship one. The result is creative and well-reasoned solutions.
Outside of robots, I read, travel, watch pretty much any movie, and love geeking out over how humans think.
The Catalyst (Capstone Project)
RESEARCH AND PROJECTS
For more details on the project click here.
Improving Policy Training for Manipulation tasks using World Models
Dexterous Transfer: Arms to Humanoids
We're building a mobile manipulator to automate labor-intensive tasks in lab environments.
On the technical side, I worked on the autonomous navigation stack. We used a 3D LiDAR with FAST-LIO for SLAM and localization.
Ground truth v/s world model hallucinated data
3D SLAM
Indoor Navigation
Sensor Fusion
LiDAR-Intertial Odometry
VLA
World Models
Robot Learning
RL PPO
Isaac Lab
Hierarchical planning
We evaluate whether data generated by world models, despite being visually bad, can still be utilized for policy training and suggest a new training objective for World Models.
Explored object-centric hierarchical planning for bi-manual dexterous manipulation on fixed arms.
We used MoCap data for the high-level wrist motion-planner and used RL finetuning for the contact-rich physics (low level planner).
Further extended the policy onto Unitree G1 humanoid.
DPSO-Q: A Reinforcement Learning–Enhanced Swarm Algorithm for Solving the Traveling Salesman Problem
DPSO-Q blends particle swarm optimization with Q-learning to solve the classic Traveling Salesman Problem more efficiently. It cuts route lengths by up to 7.5% over standard swarm methods and runs over 90% faster than ant-colony approaches.
Swarm Robotics
RL Optimization