Head of Physical AI (AI/Robotics)

Location: Remote — Europe and the United States preferred; exceptional candidates globally will be considered. Travel as needed.

Company

Builds the data, evaluation, and deployment layer for Physical AI across multimodal robot and human data, annotation, model evaluation, and physical-system deployment.

What you will do

• Establish the Physical AI research and engineering function and define a focused roadmap with hypotheses, milestones, success measures, and stop criteria.
• Design, adapt, train, and evaluate Transformer-based systems for embodied tasks, including vision-language-action models, multimodal Transformers, robot foundation models, world models, and robot policies.
• Define sensor modalities, annotations, data mixtures, quality controls, evaluation baselines, release gates, and held-out conditions.
• Take projects from problem definition through training, hardware integration, real-world validation, and staged deployment with supervision and rollback mechanisms.
• Analyze failures across data, perception, models, control, hardware, and operating environments.
• Recruit and lead a small team while communicating technical strategy, evidence, uncertainty, and limitations to customers, partners, and investors.

Requirements

• Personally made material architecture or training decisions in a substantial Transformer-based system, such as a vision, multimodal, video, world-model, planning, or control system.
• Experience representing and tokenizing inputs and outputs, fusing modalities, structuring attention and temporal context, selecting losses, building data mixtures, monitoring distributed training, and improving task performance through failure diagnosis.
• Machine-learning experience with a system that perceives or acts in the physical world, with at least one substantial project progressing beyond offline data or simulation into a real or operational physical system.
• Strong Python engineering skills, direct PyTorch or equivalent deep-learning framework experience, and the ability to debug unfamiliar model and training code.
• Experience with large multimodal datasets, controlled experiments, rollout-failure analysis, and trade-offs involving compute, memory, training stability, inference latency, and cost.
• Technical ownership of a significant research, model-development, robotics, or cross-functional program, including architecture and resource decisions and delivery into deployment.

Nice to have

• Deep computer-vision experience in representation learning, VLMs, video modeling, detection, segmentation, tracking, 3D reasoning, pose estimation, calibration, localization, sensor fusion, or real-time vision.
• Experience with vision-language-action models, robot foundation models, action tokenization, diffusion or flow-matching policies, imitation learning, reinforcement learning, world models, or cross-embodiment training.
• Experience with robot manipulation, distributed training, inference optimization, ROS 2, sim-to-real transfer, or safety-critical systems.
• Open-source work, research publications, early-stage company experience, or work with technical customers and research partners.
• PhD in machine learning, computer vision, robotics, computer science, or a related field.

Culture & Benefits

• Hands-on player-coach role with at least half of the first year focused on direct technical work.
• Hardware- and model-agnostic approach focused on measurable improvement in real tasks.
• Small, focused team with emphasis on decision quality, reproducibility, real-system results, and customer value.
• Full-time employment with travel as needed.