Builds

Tiny Worlds

Reimplemented Deepmind's Genie 1 world model from scratch to generate playable video game worlds

Humanoid Playing Piano with RL

Trained Unitree Humanoid in Mujoco to play Anime songs using Soft-Actor Critic RL and RLHF

Teleoperation and Imitation learning with Joycons and action transformer

Created a teleperation system for a bimanual robot with Nintendo Switch Joycons, trained an imitation-learning action-chunking transformer to autonomously perform household tasks. See the X thread!

Training simulated humanoid robots with reinforcement learning

Trained simulated humanoid robots with 9-actuators and RGB vision using DDPG, Decision Transformer, I2A with PPO, and SAC. See my Youtube for more!

Hierarchical Reinforcement Learning for Go2 Quadruped

Created a hierarchical reinforcement learning algorithm with a high level footstep trajectory optimizer and low level reinforcement learning joint controller to traverse extreme terrain

Shell Adaptive Kernel Density Estimation

Novel Algorithm for Kernel Density Estimation that takes advantage of low-variance query-dataset pairs to significantly accelerate the speed of adaptive random sampling. Created under the UPenn PURM grant

Speech Intelligibility Toolkit

Tools to process human audio recordings into embeddings and analyze those embeddings towards optimized inputs of speech-discerning networks. Created for the Upenn Comp linguistics lab.

GAN Engine using only matrix multiplication

GAN Engine built with no Pytorch/Tensorflow, only using NumPy, with customizable Activation functions, Cost functions, and Optimizers!