@Stanford CS ’27 | Humanoid robotics + robot learning | The Movement Lab
Stanford, CA
Joined August 2021
- Tweets7
- Following96
- Followers22
- Likes191
Pinned Tweet
Yes, GPT Astra can control a humanoid through long-horizon tasks in the real world!
HomeBody asks a simple question: can a frontier VLM skip learned VLAs entirely? @giohuh_ engineered an incredible harness that lets VLMs like Astra use persistent spatial memory to directly orchestrate composable humanoid skills — no learned VLA in the middle.
A few of my favorite moments beyond the main demo ↓
What can Astra do when given a humanoid embodiment?
We built HomeBody to find out. Controlled by GPT Astra, it carries out long-horizon tasks in a previously unseen kitchen—from tidying up across the room to retrieving remembered objects from ambiguous requests—without environment-specific training data or additional policy learning.
Here's how we did it 👀: tml.stanford.edu/homebody/
Cayden Gu retweeted
#HomeBody is soon going to be public for everyone!
Make sure you star github.com/Stanford-TML/home… and follow the updates.
@giohuh_ @CaydenGu @TakaraTruong
We’ve updated the HomeBody repo with our code release schedule for the coming weeks 📅!
Stay tuned here: github.com/Stanford-TML/home…
3/ Real robots fail — so HomeBody is built to recover and retry.
If a pick is slightly off, the underlying skill can track the target, adjust, replan, and retry locally — without asking Astra to make a new high-level decision for every correction.
If local recovery is exhausted, the failure reason is returned to Astra, which can reposition, choose a new target, or change the plan.
Here, the G1 misses the pill bottle and gets it on the retry:
What excites me most about HomeBody is upgrade path ahead. System 2 + System 0 lets the two sides improve independently, without having to retrain a monolithic VLA every time either one gets better.
Better VLMs can bring stronger spatial reasoning, planning, and faster high-level decisions. Better robot skills can bring more reliable physical execution across a wider range of tasks.
As both improve, the interface between them stays simple. That’s the direction I’m most excited to explore with future works/demos with HomeBody!
Cayden Gu retweeted
What can Astra do when given a humanoid embodiment?
We built HomeBody to find out. Controlled by GPT Astra, it carries out long-horizon tasks in a previously unseen kitchen—from tidying up across the room to retrieving remembered objects from ambiguous requests—without environment-specific training data or additional policy learning.
Here's how we did it 👀: tml.stanford.edu/homebody/