Kartikay Milind Pangaonkar
kpangaon@asu.edu

I'm a robotics engineer pursuing my Master's degree at Arizona State University, focused on learning-based control and policy generation for real-world robotic systems. I have spent the last two years researching in robot learning, with hands-on experience developing reinforcement learning and diffusion-based policies and deploying them from simulation to physical platforms.

I am comfortable working across the full robotics stack — including perception, state estimation, and control — with strong experience in ROS, Python, and GPU-accelerated ML.

I'm always interested in building with foundation models like VLAs and World Models to develop robotic systems that operate reliably outside controlled environments. Feel free to reach out!

picture_as_pdf CV school Google Scholar business LinkedIn

Profile Picture

News

Jan '26 🎉 FDP full paper accepted at ICRA 2026.
Aug '24 🎓 Started Master's in Robotics and Autonomous Systems at Arizona State University.
Jun '25 🎉 FDP abstract paper accepted at RSS 2025 RoboReps Workshop.
Early '26 📄 StageCraft paper submitted — training-free VLA improvement via VLM-based environment staging.

Research

My research focuses on learning-based control and policy generation for robotic systems, with emphasis on making policies that generalize across diverse environments and tasks. I work on reinforcement learning, diffusion-based policies, and foundation model-based approaches for robust real-world robot deployment. Here are my recent publications:

 StageCraft: Execution Aware Mitigation of Distractor and Obstruction Failures in VLA Models
Kartikay Milind Pangaonkar*, Prabin Kumar Rath*, Omkar Patil*, Nakul Gopalan
Under Submission, 2026
We propose a training-free approach to improve pretrained VLA policy performance by manipulating the environment's initial state using VLM-based in-context reasoning. We evaluate performance of state-of-the-art VLA models with StageCraft and show an absolute 40% performance improvement across three real world task domains involving diverse distractors and obstructions.
picture_as_pdf Paper language Website
Factorizing Diffusion Policies for Observation Modality Prioritization
Omkar Patil, Prabin Kumar Rath, Kartikay Milind Pangaonkar, Eric Rosen, Nakul Gopalan
International Conference on Robotics and Automation (ICRA), 2026
RSS RoboReps Workshop, 2025
We present a theoretical framework to learn action diffusion models without the need to jointly condition on all input modalities. Our method is robust to deploy against visual distractors and appearance changes, maintaining strong performance even under significant visual disruptions and outperforming standard diffusion policies by over 40%.
picture_as_pdf Paper language Website

Projects

Goal Conditioned RL for Tabletop Manipulation
Arizona State University, 2025
Implemented a goal-conditioned reinforcement learning framework for multi-object robotic manipulation using an Entity Interaction Transformer (EIT) with object-centric visual representations derived from Deep Latent Particles. Trained the agent using TD3 with Hindsight Experience Replay in simulated IsaacGym environments, enabling it to learn from pixel observations and generalize across varying numbers of objects.
PPO-Based Framework for Autonomous Driving in AWS DeepRacer
Arizona State University, 2025
Designed and trained a reinforcement learning agent for autonomous driving using AWS DeepRacer. The agent learns to navigate race tracks efficiently and safely using Proximal Policy Optimization (PPO).
Improving Transformer Efficiency Using AdaLN-Zero for Robot Manipulation
Arizona State University, 2025
Implemented and improved a diffusion-based visuomotor policy for robotic manipulation by reproducing the Push-T task from Diffusion Policy (Chi et al., 2024). Proposed replacing cross-attention conditioning with Adaptive Layer Normalization Zero (AdaLN-Zero) to reduce computational overhead and improve transformer efficiency.
Balancing of a Two-Wheeled Segway Robot
Arizona State University, 2024
Modeled and stabilized a two-wheeled self-balancing Segway robot using modern control theory by formulating a state-space representation as an inverted pendulum. Analyzed controllability, observability, and stability; designed a state-feedback controller using pole placement; validated results through MATLAB simulations demonstrating stable closed-loop behavior.
Design of Thrust Controller for Propeller-Based UAVs
Arizona State University, 2024
Developed a thrust controller for propeller-based UAVs to maintain constant thrust despite changes in air density with altitude. Designed a PID controller in Simulink to adjust motor RPM based on thrust error. Propeller dynamics are modeled using Blade Element Theory and validated through CFD simulations.
Design and Fabrication of Peltier-Based Thermoelectric Battery Cooling System
College of Engineering Pune, 2024
Designed an intelligent battery thermal management system that regulates battery temperature in two-wheeler electric vehicles. The system leverages the thermoelectric effect of Peltier modules to reject heat into liquid heat sinks and the surrounding environment.