Embodied AI / Simulation Engineer

About the team:

The Software Engineering Team builds and operates the systems that power Formic’s Robotics-as-a-Service platform. The team develops the monitoring, provisioning, and data infrastructure that enables real-time visibility across a growing fleet of deployed industrial robotic systems. Engineering ensures systems are observable, resilient, and remotely diagnosable at scale by building production-grade edge and cloud systems that support reliable data collection, remote troubleshooting, live video streaming, and continuous system improvement. This work directly impacts fleet uptime, service efficiency, and customer outcomes.

Within Engineering, Formic is launching a brand-new Advanced Robotics Team focused on building next-generation intelligent robotic systems designed to operate in real industrial environments. This team will develop wheeled humanoids and advanced mobile manipulation platforms, pushing beyond classical robotics stacks into modern embodied AI systems.

Formic is uniquely positioned to do this. We already operate hundreds of deployed robotic systems and maintain production-grade deployment, monitoring, and fleet infrastructure across real manufacturing customers. This provides what most humanoid startups do not: continuous real-world data, production testing environments, and a clear path from research to scaled deployment.

This is not a demo lab. We are building learning-driven robotic systems that operate daily in real factories.

About this role:

As an Embodied AI / Simulation Engineer, you will develop learning-based manipulation systems for humanoid and mobile manipulation platforms and ensure they transfer reliably into the real world. You will build the simulation, data, and training infrastructure needed to develop robust visuomotor policies and deploy them onto physical robotic systems operating in production environments.

This role sits at the intersection of simulation, machine learning, and robotics execution. You will work closely with perception and controls teams to ensure learned policies operate safely, reliably, and effectively in closed-loop real-world conditions.

In this role you will:

• Design and train learning-based manipulation systems for humanoid and mobile manipulation platforms
• Develop and maintain high-fidelity digital twins using Isaac Sim, MuJoCo, or similar simulation tools
• Implement and evaluate approaches including Action Chunking with Transformers (ACT), diffusion policies, behavior cloning, and Vision-Language-Action (VLA) models
• Contribute to development of a Universal Manipulation Interface (UMI) abstraction layer
• Build teleoperation-to-training data pipelines to enable scalable dataset generation
• Design sim-to-real transfer strategies including domain randomization and system identification
• Evaluate policy robustness, failure modes, and cross-task generalization
• Partner with perception and controls teams to ensure stable closed-loop visuomotor policies
• Deploy trained models onto real robotic systems and support on-hardware validation and debugging
• Establish experimental rigor through structured evaluation, ablations, and performance tracking

What makes you a great fit:

• Experience training embodied AI policies for real robotic systems
• Strong understanding of sim-to-real challenges and real-world failure modes
• Familiarity with transformer-based action models such as ACT
• Experience with diffusion policies or other generative control approaches
• Experience working with multi-modal inputs including vision, proprioception, and language
• Proficiency in Python and deep learning frameworks such as PyTorch or JAX
• Experience integrating learned policies with real-time control system
• Strong experimental design, evaluation discipline, and attention to reproducibility
• Ability to operate across research and deployment constraints without losing rigor