M

Robotic AI Engineer/Applied Scientist - Foundation Models

Maven Robotics
2 hours ago
Full-time
On-site
San Francisco, California, United States
Company Overview

Maven Robotics

is building the world's leading general-purpose robots and providing physical AI solutions for the most challenging industrial autonomy tasks.

Operating in stealth, we are assembling a team of world-class innovators who think from first principles. Our mission is to achieve human-level task success rates in complex environments, even when faced with limited fine-tuning data or evolving robotic hardware. We value unwavering truth-seeking, humility, and relentless determination. Role Description

We are seeking exceptional AI researchers and engineers to architect the neural backbone of our general-purpose robots. You will design the

Vision-Language-Action (VLA)

frameworks or

Action World Models

that allow our robots to reason, adapt, and succeed where traditional automation fails.

Note on Leveling:

We are hiring across

all levels

(from early-career to staff/principal). You are not expected to be a master of every domain listed below; however, you must be able to justify

world-class excellence in at least one core factor

(e.g., model architecture, RL formulations, or high-scale data systems).

In this role, you will: Architect Embodied Foundations:

Design model architectures (VLA, World Models, etc.) that achieve ultra-high task success rates with minimal human demonstrations. Master Data Efficiency:

Develop novel co-training strategies and efficient learning algorithms that leverage diverse data sources-from Internet-scale video to sparse, high-fidelity human interventions. Generalize Cross Embodiments:

Build models capable of zero-shot or few-shot adaptation to new robot configurations, maintaining a high success rate even when proprietary hardware and actuation systems evolve. Innovate Real-World RL:

Formulate and deploy novel Reinforcement Learning and policy extraction methods specifically designed for physical, real-world manipulation. Design the Data Loop:

Collaborate on advanced data collection systems to capture critical human intervention data for model bootstrapping. Qualifications

Must-have: MS or PhD

in CS, Robotics, Machine Learning, or a related field (or equivalent industry experience). Deep Technical Mastery:

Advanced understanding of transformers, multi-modal alignment, and mapping perception to high-frequency motor control. Specialized Excellence:

Proven ability to innovate-not just implement-within one or more areas: VLA models, Real-world RL, or large-scale Data Infrastructure. Software Excellence:

Expert-level Python and deep familiarity with PyTorch or JAX Nice-to-have:

A track record of high-impact publications (NeurIPS, ICRA, RSS, CVPR) or significant open-source contributions. Experience with large-scale distributed training and model compression. Experience with deployment of models to edge devices (NVIDIA Jetson/Orin) for real-time inference. General knowledge of robotics principles (kinematics, dynamics). Why Maven Robotics?

You will be working on a unique in-house robot platform that is evolving alongside your models. We don't just want to reach 90% success; we are building the architectural breakthroughs required to reach 99.9% in the real world.