Skip to content
View edwhu's full-sized avatar
🔍
🔍
106 followers · 0 following

Organizations

@ctcusc@clvrai@DeepSymphony@penn-pal-lab

Block or report edwhu

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
edwhu/README.md

Hi, I'm Edward. Currently, I'm interested in developing data-driven methods that interact, explore, and learn from the world. My research investigates deep reinforcement learning, perception, and robotics

🤖 Research Artifacts

Here are the codebases of my research projects so far.

open-dreamerFull-scale reimplementation of Dreamer-4
aawrReal World RL of Active Perception Behaviors (NeurIPS'25)
dreamer4-jaxPrototype reimplementation of Dreamer 4 in Jax
bstThe Belief State Transformer (ICLR'25)
scaffolderPrivileged Sensing Scaffolds RL (ICLR'24 Spotlight)
planning goals for explorationPlanning Goals for Exploration (ICLR'23 Spotlight)
interactive reward functionsTraining Robots to Evaluate Robots (CoRL'22 Best Paper Award)
robot aware controlKnow Thyself: Transferable Visual Control Policies Through Robot-Awareness (ICLR'22)
ikea furniture simulationIKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks (ICRA'21)

Nerd stuff

I like studying codebases that are elegantly written and do cool things. Some topics that I found interesting lately: dataloading at scale, Jax renderer, Jax Monte-Carlo Tree Search library.
Some of my side projects:

gpt2 slackbotChatting with GPT2 in slack
optical illusionA cool optical illusion
Something to think about as an ML researcher I think researchers start out very pure hearted, but can easily end up misled and lost. The incentives of the modern research community, particularly ML, are misaligned with doing good science. To employ an analogy, ML is currently like a hackathon. You are incentivized to put together an MVP that works just enough to pass the appraisal of the judges. You feel obligated to use the shiny new "X" because it will garner public attention. Companies with free t-shirts and kickbacks swarm around you.

Yes, some of these things are unavoidable. But if you blindly follow the noise, you may end up in the eye of the storm - at a standstill, with no exit in sight.

Ant Death Spiral

Don't be distracted by the noise!

Pinned Loading

  1. clvrai/furnitureclvrai/furniturePublic

    IKEA Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks

    Python 573 63

  2. penn-pal-lab/pegpenn-pal-lab/pegPublic

    Code for "Planning Goals for Exploration", ICLR2023 Spotlight. An unsupervised RL agent for hard exploration tasks.

    Python 83 6

  3. penn-pal-lab/interactive_reward_functionspenn-pal-lab/interactive_reward_functionsPublic

    Code release for "Training Robots to Evaluate Robots" (CoRL'22, Best Paper Award)

    Python 17 1

  4. penn-pal-lab/scaffolderpenn-pal-lab/scaffolderPublic

    Official codebase for "Privileged Sensing Scaffolds Reinforcement Learning", contains the Scaffolder algorithm and Sensory Scaffolding Suite.

    Python 33 3