Sandeep Bajamahal

I'm a second-year EECS undergraduate student at UC Berkeley, with interests in reinforcement learning, deep learning, and robotics. I'm also an undergrad researcher at Ken Goldberg's AUTOLab, where I focus on generalist robot policies. Previously, I attended Saint Francis High School, during which I led FRC Team 2367's code team and directed robotics at Allstars Robotics.

Email / Linkedin / Personal Website / Github

Research

I’m interested in researching machine learning for robotics and self-driving, with a focus on understanding and improving the performance of VLA foundation models.

OXE++: Scaling Robot Augmentation for Cross-Embodiment Policy Learning
June 2022
Guanhua Ji*, Harsha Polavaram*, Lawrence Yunliang Chen*, Sandeep Bajamahal, Zehan Ma, Simeon Adebola, Chenfeng Xu, Ken Goldberg
[under review]

We study how scaling robot augmentation impacts cross-embodiment learning and introduce OXE++, a large-scale augmented version of the Open X-Embodiment dataset. Augmenting demonstrations across robot embodiments improves policy generalization to unseen robots, achieving 24–45% higher success on real-world manipulation tasks.

Smart Path Generation using Model Predictive Control
August 2023
[paper]

Investigated the application of model predictive control(MPC) in path planning and object avoidance. I simulated autonomous robot movement around stationary obstacles using MPC and Pure Pursuit.

Projects

	Learning to Walk with Reinforcement Learning January 2025 [code] [video] I implemented and trained a Soft Actor Critic (SAC) network to learn to walk and beat OpenAI's BipedalWalker environment. Trained for ~3 hours on my M1 Mac, achieving a final reward of 306. (Solve score is 300)
	Reinforcement Learning Pong Agent November 2024 [code] I wrote two iterations of teaching a RL neural network to play pong. In the first volatile iteration I had the agent train after receiving a instant reward. In the optimized iteration I trained the agent to maximize expected reward.
	Writing a Neural Network from Scratch November 2024 [code] A scalable neural network class written with only Numpy. I wrote forward/back propagation and gradient descent from scratch.
	Facial Expression Music Recommender Fazal Mittu, Ruhi Yusuf, Sandeep Bajamahal, Koena Jaware, Paree Merchant, Siddhant Hullur June 2022 [blog] [code] I created a dataset of Spotify songs based on emotion and then trained a classifier to infer the emotion of new songs provided by a user's playlist.

Writing

I've enjoyed writing technical blogs on my Medium profile.

An Introduction to GraphSLAM
October 2024
[blog] [code]

An introduction and explanation of the mathematics behind solving graphSLAM. I also included a Python program implementation.

Writing A Custom Robot Control System
June 2023
[blog] [code]

I wrote a Python robot control system using UDP and threading to wirelessly control up to 4 Raspberry Pi robots. Both server and robot-side code are included. This project is a cost-effective version of FRC Driverstation that I wrote for my high school robotics team.

Site design from Jon Barron