VERTEX Contact
VERTEX / REMOTE · UNITED STATES / EST. 2026

Foundation models
for extreme-purpose
robotics.

Foundation models for firefighters, combat medics, hazmat teams, search-and-rescue crews, and the soldiers who go down range. We build the model layer for robots that work where every additional second of human presence is a tradeoff against death.

Open roles Read the research

§ 01 — THE THESIS

Generalist physical intelligence for the people doing the hardest work in the world.

The hardest jobs in the world.

§ 02 — CAPABILITIES

What we’re building.

01—06

01

Whole-body control

Unified policies for locomotion, balance, and manipulation in unstructured environments — collapsed buildings, smoke, debris, contested ground.

Target: real-time closed-loop control

02

Cross-embodiment transfer

One backbone, many platforms. We’re designing for adapter-based transfer so a new chassis can be brought online in hours, not months.

Goal: minutes-scale adaptation

03

Vision-language-action

Operators give natural-language goals; the model grounds them in scene geometry and proprioceptive state. No bespoke programming per mission.

Multimodal: vision · language · proprio

04

Sim-to-real

Domain-randomized rollouts close the reality gap before a single real-world episode. Critical for environments that are too dangerous to train in directly.

Distributed simulation pipeline

05

Long-horizon planning

Hierarchical agents decompose multi-step missions — search, extraction, triage — into recoverable sub-goals with continuous human supervision.

Multi-step task decomposition

06

Safety & oversight

Every deployed policy is designed to ship behind a learned safety layer and a real-time human-override channel. Operators stay in command.

Human-in-the-loop by default

§ 03 — RESEARCH

Research that moves
off the page.

We work across the model, data, and evaluation layers required to turn generalist physical intelligence into reliable systems in the field.

01 / PUBLICATIONS

Generalist policies for the physical world.

Research across whole-body control, multimodal reasoning, adaptation, and long-horizon autonomy.

View publications 02 / OPEN SOURCE

Infrastructure others can inspect and build on.

Code, tools, and reference implementations for the systems we can responsibly release.

Explore open source 03 / ROADMAP

A clear line from research to deployment.

What we are building now, how we evaluate it, and the milestones between a model and field readiness.

Read the roadmap

§ 04 — OPERATORS

Who we build for.

All four, side by side — the operators hub
Structure fire — operator photograph plate STRUCTURE FIRE → For the firefighters. Prehospital and TCCC — operator photograph plate PREHOSPITAL / TCCC → For the medics. Urban search and rescue — operator photograph plate USAR · WILDERNESS → For the rescue crews. Reconnaissance and CASEVAC — operator photograph plate RECON · CASEVAC → For the military.
Romir Patel — scratchboard portrait
01 PORTRAIT · SCRATCHBOARD

§ 05 — TEAM

Founded by a researcher who has shipped in this field.

Romir Patel Founder & CEO

prev. MIT CSAIL, Caltech, Bespoke Labs

§ 06 — CAREERS

Join us.

We’re a small, fully remote team across the United States. We work synchronously Monday through Thursday, with Friday reserved as a think day. We hire researchers and engineers with strong opinions and a track record of finishing things.

RESEARCH Research Scientist, Foundation Models United States RESEARCH Research Engineer, Reinforcement Learning United States RESEARCH Research Engineer, Evaluation & Benchmarks United States GO-TO-MARKET Go-to-Market United States
See all open roles