Rohan Tan Bhowmik bio photo

Email

My CV

LinkedIn

Github

Google Scholar

ResearchGate

Hello! Welcome to Rohan Tan Bhowmik’s website. I am a senior at Stanford University pursuing Electrical Engineering and Computer Science with a focus on artificial intelligence.

I am a researcher at the Intelligence through Robotic Interaction at Scale (IRIS) Lab under Professor Chelsea Finn, where I teach humanoid robots to play sports and compete against humans. I recently served as co-president of Stanford Student Robotics and as software lead for RoboSub, Stanford’s autonomous underwater vehicle project.

I have hands-on experience applying AI to robotics, health, and human-computer interaction, as well as with machine learning compilers for optimizing foundation model performance on heterogeneous computing architectures.

Academic Background

  • Sep. 2023 - Present: Stanford University
    • B.S., Electrical Engineering (Class of 2027)
    • M.S., Computer Science (expected Dec. 2027)
  • Aug. 2016 - Jun. 2023: The Harker School

Experience

Researcher, Stanford IRIS Lab

  • Human-out-of-the-loop Robot Learning (Jun. 2026 - Present). Developing a task-and-scene proposal pipeline for autonomous self-play, enabling robot policies to acquire diverse manipulation skills with minimal human supervision. Designing a framework for continuous skill acquisition — day-time autonomous exploration and reset, night-time trajectory-based training — across a multi-arm setup.
  • Correction and Reinforcement from Evidence (May 2026 - Present). Developed a data collection strategy prioritizing failure-distribution coverage and counterfactual demonstrations to optimize policy training efficiency, and designed a dual-signal training loop combining human corrections and policy rollouts to jointly update a policy actor and an Implicit Q-Learning (IQL) critic.
  • Learning in Partially Observable Environments (Oct. 2025 - Present). Investigated policy training under partial observability, addressing self-occlusion failure modes such as an arm blocking the camera line-of-sight during manipulation.

Independent Research

  • Humanoid Robots (Jan. 2026 - Present). Training a Unitree G1 humanoid to play table tennis at a competitive level via human demonstration and reinforcement learning with sim-to-real transfer, alongside other sport and movement skills including football throwing, soccer kicking, and dancing.
  • Spiking Neural Networks (Apr. 2026 - Present). Extended Hodgkin-Huxley neuron dynamics to model biological neurogenesis, enabling a spiking neural network to grow and prune neurons over time as it learns. Cut catastrophic forgetting by ~3x versus a fixed-size network by protecting mature neurons during continual learning, validated via an ablation study.
  • Quantum Optical Convolutional Neural Networks (Jun. 2019 - Mar. 2020). Created a novel framework to simulate the training of convolutional neural networks on quantum computing systems, demonstrating >100x boosts in computational and memory efficiency over a classical CNN.

Leadership and Industry

  • Co-President, Stanford Student Robotics Club (Sep. 2025 - Jun. 2026). Led student engagement with robotics on campus by organizing project showcases, company events, and hackathons, growing participation and industry partnerships across the club.
  • Software Lead, Stanford Autonomous Underwater Vehicle Project (Oct. 2024 - Jun. 2026). Led the design and implementation of a ROS 2 based autonomy stack integrating perception, localization, and path-planning, and managed a cross-disciplinary software team of ~15 engineers delivering a field-validated robotics platform for competitions and scientific missions.
  • Software Engineering Intern, AMD AI Group (Jun. 2024 - Sep. 2025). Specialized in machine learning compilers, optimizing model performance across diverse hardware architectures and reducing latency in large-scale deployments. Contributed to Wave, a GPU domain-specific language, by designing and extending custom kernels.

Research Intern, Stanford University

  • California Wildfire Prediction System (Jun. 2022 - Jan. 2024). Developed a wildfire prediction system based on a novel machine-learning framework, successfully predicting 85% of major wildfires in California up to 2 weeks in advance, and compiled a comprehensive wildfire dataset with >37 million data points.
  • AI-assisted Facial Surgery Planning and Evaluation (Jun. 2021 - May 2022). Developed AI algorithms to aid surgical planning for correcting facial structural deformities that cause debilitating breathing, sleeping, and sensory disorders, achieving >95% accuracy in classifying patient conditions.
  • COPD Exacerbation Prediction (Jun. 2020 - May 2021). Developed a multi-modal solution for predicting the climate’s impact on exacerbation risks of respiratory diseases, built on a new architecture combining convolutional and recurrent neural networks with 93% cough detection accuracy.

Research Interests

I am constantly seeking to learn and develop new machine-learning techniques and build applications based on them, especially in the areas of robotics, health, environment, and human-computer interaction. I am deeply interested in brain-inspired computational architectures for embodied intelligence.

Contact

You may find me at rohan[dot]t[dot]bhowmik[at]gmail[dot]com or rbhowmik[at]stanford[dot]edu.