---
title: "Research Scientist, Vision-Language-Action Foundation Models"
slug: "research-scientist-vision-language-action"
id: "119b09e1-7cb2-498c-9c43-3ffcfef4ec05"
department: "Robotics & Embodied AI"
department_code: "ROB"
location: "San Francisco, CA"
workplace_type: "On-site"
employment_type: "Full-time"
compensation: "$265k – $375k (annual package)"
is_remote: false
is_featured: false
headcount: 1
canonical_url: "https://jobs.talent247.co/jobs/research-scientist-vision-language-action"
markdown_url: "https://jobs.talent247.co/jobs/research-scientist-vision-language-action.md"
apply_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-vision-language-action/apply"
schema_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-vision-language-action/form-schema"
webmcp_tool: "apply_for_role"
updated_at: "2026-11-15T17:34:20.564Z"
content_signal: "search=yes, ai-input=yes, ai-train=no"
---

# Research Scientist, Vision-Language-Action Foundation Models

## Position Telemetry

- **Reference:** `RESEARCH-SCIENTIST`
- **Practice / Sector:** Robotics & Embodied AI (`ROB`)
- **Location:** San Francisco, CA
- **Workplace Type:** On-site
- **Employment Type:** Full-time
- **Compensation:** $265k – $375k (annual package)
- **Canonical HTML URL:** https://jobs.talent247.co/jobs/research-scientist-vision-language-action

## Role Dossier & Specification

THE CLIENT & MANDATE
Our client is one of the world's most closely watched physical-intelligence research laboratories, headquartered in San Francisco and backed by top-tier Silicon Valley venture investors and AI pioneers. Their mission is to bring general-purpose foundation models into the physical world—training unified Vision-Language-Action (VLA) models capable of controlling diverse robotic morphologies across dexterous, deformable-object, and long-horizon real-world tasks.

They have retained Talent247 to appoint a Research Scientist specializing in Vision-Language-Action Foundation Models onsite in San Francisco. Working alongside co-founders and robot fleet teams, the successful candidate will advance pre-training and post-training architectures that bridge internet-scale semantic reasoning with continuous, high-frequency dexterous motor control.

KEY RESPONSIBILITIES
• Design and scale generalist Vision-Language-Action architectures uniting pretrained multimodal backbones with continuous flow-matching, diffusion, or autoregressive action-token heads.
• Investigate cross-embodiment transfer representations enabling a single foundation policy to share physical intuition across single-arm, bimanual, mobile-manipulator, and dexterous-hand hardware.
• Develop hierarchical planning-and-control paradigms where high-level semantic subgoal reasoning steers reactive 50Hz+ visuomotor execution and autonomous error recovery.
• Pioneer post-training alignment and offline-to-online reinforcement learning recipes that turn imperfect teleoperation demonstrations into superhuman cycle speed and 99%+ task completion reliability.
• Conduct rigorous scaling-law studies across heterogeneous data mixtures combining own-fleet teleoperation, open robotics corpora, egocentric human video, and synthetic simulation.
• Validate every algorithmic advance directly on an expansive in-house fleet of physical robots operating continuously on site.

TECHNICAL ENVIRONMENT & OFFERING
Stack: Python, JAX, PyTorch, flow-matching/diffusion action transformers, multi-node H100/B200 training clusters, large-scale in-house bimanual & mobile robot fleet. Workplace: Onsite in San Francisco, CA. Compensation: $265,000 – $375,000 base salary plus premier early-stage research lab equity and comprehensive benefits.

## Core Requirements

ESSENTIAL QUALIFICATIONS
• PhD in Robotics, Machine Learning, Computer Vision, or equivalent frontier research track record in robot learning and multimodal foundation models.
• First-author publications at RSS, CoRL, ICRA, NeurIPS, ICML, ICLR, or CVPR in visuomotor policy learning, VLA models, imitation learning, or generative control.
• Expert proficiency in JAX or PyTorch across multi-node distributed training, custom attention masking, and continuous generative modeling (flow matching / diffusion).
• Proven hands-on experience deploying learned neural policies onto real physical robot hardware—not purely simulated benchmarks.
• Deep intuition for failure-mode debugging across camera placement, action chunking, proprioceptive conditioning, and data distribution shift.

PREFERRED / BONUS EXPERIENCE
• Prior core contributions to landmark generalist robot policies, large-scale cross-embodiment datasets, or dexterous manipulation breakthroughs.
• Experience combining VLM pre-training or egocentric video world models with downstream robotic control fine-tuning.
• Strong systems mindset capable of co-designing model architectures around real-time 20ms–50ms onboard control-loop constraints.

## Application Schema & Direct Apply (`apply_for_role`)

Candidates and authorized AI agents on behalf of a candidate can submit an application via the declarative WebMCP form tool `apply_for_role` on https://jobs.talent247.co/jobs/research-scientist-vision-language-action or via HTTP `POST https://jobs.talent247.co/api/public/jobs/research-scientist-vision-language-action/apply` (`application/json` or `multipart/form-data`). Non-browser HTTP agents must include header `X-Talent247-Intake-Token` obtained from `GET https://jobs.talent247.co/api/public/intake-token` (or automatically returned in `GET https://jobs.talent247.co/api/public/jobs/research-scientist-vision-language-action/form-schema`).

### Form Fields

- `full_name` (text, required): Full name
- `email` (email, required): Email
- `phone` (phone, optional): Phone
- `location` (text, optional): Location
- `linkedin_url` (url, optional): LinkedIn URL
- `cover_letter` (textarea, optional): Cover letter
- `consent` (checkbox, required): I consent to my data being processed for this application
- `cv_file` / `cv_base64` + `cv_filename`: Upload a CV in .pdf or .docx format (<= 10 MB) or provide an `https://` LinkedIn profile URL in `linkedin_url`.

### Alternative: Talent Pool Intake (`join_talent_pool`)

If this mandate is not an exact match, register once in the Talent247 Talent Pool at https://jobs.talent247.co/talent-pool (`https://jobs.talent247.co/talent-pool.md`) via WebMCP tool `join_talent_pool` or `POST https://jobs.talent247.co/api/public/talent-pool`.
