Head of Physical AI

This is a fully remote position, open to applicants in United States.

📋 Description

• Establish and oversee the AI research and engineering division.

• Determine the Physical AI challenges the company will tackle and establish evaluation criteria for results.

• Stay actively engaged in model design and experimentation, code writing or review, data inspection, training run debugging, failure analysis, and robot rollout assessments.

• Create a targeted Physical AI research and engineering roadmap that includes hypotheses, baselines, milestones, success metrics, and criteria for stopping projects.

• Choose model families for training, adaptation, or evaluation and make decisions about building internally, utilizing open models, licensing technology, or collaborating with partners.

• Design, modify, train, and assess Transformer-based systems for embodied tasks.

• Work with multimodal and embodied data, vision-language-action models, multimodal Transformers, robot foundation models, action representation, imitation learning, reinforcement learning, world models, cross-embodiment transfer, and robot-policy assessment.

• Specify sensors, modalities, annotations, data mixtures, coverage, and quality control measures.

• Evaluate how data quality, diversity, and composition influence model performance and real-world task execution.

• Create reproducible offline and real-world evaluation systems, including baselines, held-out conditions, and release criteria.

• Safeguard evaluations against data leakage, overfitting, and ineffective or misleading success criteria.

• Advance projects from problem definition through training, hardware integration, and real-world validation.

• Investigate failures related to data, perception, models, control, hardware, and the operating environment.

• Develop safe, staged physical testing and deployment strategies with supervision and rollback options.

• Recruit and manage a small team of researchers and engineers.

• Oversee architecture, experimentation, code, rollout, and failure evaluations.

• Transform customer and partner requirements into actionable technical specifications.

• Effectively communicate technical strategy, evidence, uncertainties, and limitations to customers, partners, and investors.

• Within 90 days: audit data, evaluation assets, partnerships, and model opportunities; select research priorities; create a baseline and evaluation suite; define physical validation; present a 12-month roadmap.

• Within six months: train, adapt, or evaluate a Transformer-based model or robot policy; conduct tests on a real robot; document any failures.

• Within twelve months: showcase measurable improvements; close the model/data/retraining/evaluation loop; deliver a usable model, benchmark, evaluation system, or deployment; build a capable small team.

⛳️ Requirements

• Extensive hands-on expertise with Transformers; have made significant architectural or training decisions in at least one major Transformer-based system.

• Experience in machine learning for systems that operate in the physical world.

• At least one substantial project that has progressed beyond offline datasets or simulations into an actual operational physical system.

• Simulation experience is acceptable only when combined with credible sim-to-real ownership and physical validation.

• Strong engineering skills in Python.

• Direct experience with PyTorch or a comparable deep-learning framework.

• Ability to read and debug unfamiliar model and training code.

• Experience in designing controlled experiments and analyzing rollout failures.

• Experience handling large multimodal datasets.

• Sound judgment regarding compute, memory, training stability, inference latency, and cost considerations.

• Experience in providing direction for significant research, model development, robotics, or cross-functional technical initiatives.

• Experience making architectural and resource decisions, mentoring or hiring technical talent, halting unproductive lines of work, and assisting in deploying results.

• A management title is not necessary, but direct technical ownership is required.

• Strongly preferred: in-depth Transformer expertise paired with computer vision experience.

• Valuable experience includes vision-language-action models, robot foundation models, action tokenization, continuous action generation, diffusion or flow-matching policies, imitation learning, reinforcement learning, world models, cross-embodiment training, robot manipulation, distributed training, inference optimization, ROS 2, sim-to-real transfer, safety-critical systems, open-source contributions or research publications, early-stage company engagement, and interactions with technical customers or research partners.

• A PhD is beneficial but not mandatory; equivalent original model work, technical leadership, and real-system delivery experience are equally pertinent.

🏝️ Benefits

• Full-time employment.

• Preference for remote work in Europe and the United States; exceptional candidates from other locations will be considered.

• Travel as necessary.