Oliver Groth

@omgroth

Principal AI Engineer @vesoma_ai; prev: RS @GoogleDeepMind; research @oxfordrobots @Oxford_VGG @StanfordSVL AI researcher, robot trainer and video game nerd.

Munich, Germany

ogroth.github.io

Joined July 2018

567 Following

2,854 Followers

Replies Reposts Media

Pinned
Oliver Groth
@omgroth
Sep 14
Development sneak peek: Our boxy prototype “Kyle” already walks with confidence. 🤖 A testament to what a dedicated team with full ownership of HW and AI stack can build from scratch in just a few months. Very excited to be part of this amazing crew moving fast together. 🚀 1/2
Nikolai Ensslen
@nikensslen
Sep 14
I have watched a lot of humanoids walk over the last years. In my honest assessment, this is one of the better gaits out there - and it is still our very early stage. Congrats Martin Riedmiller & team! We decided showing the alpha platform unblurred for the first time. Until now
00:00
1
Oliver Groth
@omgroth
Sep 23
"An AI lab that ships a humanoid." That's the way to push the frontier of embodied intelligence. Very excited to be on board of this rocket ship. 🚀
Nikolai Ensslen
@nikensslen
Sep 23
Vesoma. Today the name is public. Ve for versatile, soma, Greek for body. Then we noticed the same Ve opens vera (truth) and veda (knowledge). Trustworthy, intelligent, versatile embodiments. An AI lab that ships a humanoid. Its intelligence learns from doing the work, not from
Oliver Groth
@omgroth
Sep 23
Today, we are leaving stealth mode at Vesoma: vesoma.com/news/introduci… Our vision: We want to build and ship the most versatile and trustworthy humanoid to give AI the body it needs to give us a helping hand and unblock the bottleneck of physical labour. 🧵1/6
Introducing Vesoma
From vesoma.com
7
Oliver Groth
@omgroth
Sep 1
One of the 'side-effects' of our approach was that the VLM-orchestrator could be used for one-shot, in-context learning to bootstrap RL post-training with a feasible solution. The latest releases of @SkildAI's S1 and @GeneralistAI's GEN 1.5 show how powerful ICL gets at scale.🚀
Bhavya Sukhija
@sukhijabhavy
Sep 1
Replying to @sukhijabhavy
When we started this work one year ago, there were no in-context-capable VLAs, yet. Instead VLM-orchestration for semantic exploration was our approach to go the 0->1 step, before RL can take the policy to 100. [3/4]
00:00
3
Oliver Groth
@omgroth
Sep 1
One of the best perks of my time @GoogleDeepMind was to host exceptional student researchers like @sukhijabhavy. His work on VLM-orchestrated RL for VLA post-training (done in 2025!) was way ahead of the curve and is now finally on arxiv. All details in 🧵👇
Bhavya Sukhija
@sukhijabhavy
Sep 1
New paper: EXIMO 🚀 — Explore, Imitate, Optimize How can we enable sample-efficient learning of new tasks? [1/4] Our idea: leverage the general knowledge of state-of-the-art VLMs to guide exploration toward what actually matters. 👇🧵