AI Training Infrastructure Engineer - Humanoid Whole Body Control

About the role

Figure develops general-purpose humanoid robots designed for both residential and commercial environments. We are seeking an engineer to manage the training and deployment architecture for our reinforcement learning-based whole-body control systems. In this capacity, you will own the end-to-end stack that transforms high-fidelity simulation into reliable hardware behavior. You will bridge the gap between research experiments and production-grade infrastructure, ensuring that every iteration is reproducible and measurable. Your work will directly influence how quickly control policies evolve and how safely they operate in physical settings. You will collaborate closely with controls engineers, researchers, and hardware teams to align simulation with real-world constraints. This role demands comfort with ambiguity and the ability to design systems that serve a wide range of technical users. Ultimately, you will be responsible for making the training of humanoid behaviors fast, robust, and scalable.

Key facts

What you'll do

• Architect and maintain the core infrastructure for training whole-body control policies, including data ingestion, orchestration, simulation scheduling, and telemetry visualization.
• Build configurable and reliable pipelines that integrate physics engines, asset definitions, and environment parameters to support the daily workflows of the controls engineering team.
• Drive cluster operations to maximize uptime and utilization, implementing monitoring, alerting, and recovery mechanisms that keep training cycles moving efficiently.
• Evaluate and integrate physics simulation platforms such as PyBullet, NVIDIA PhysX, Warp, or MuJoCo to optimize the balance between training throughput and physical realism.
• Design and deploy tools for hyperparameter search, experiment tracking, and result analysis that increase the efficiency of training runs and the quality of resulting policies.
• Implement interfaces and automation that carry policies from simulated training and validation stages through to deployment on real humanoid hardware.
• Partner with controls engineers to model photorealistic simulation environments and accurate contact interactions required for complex manipulation tasks.
• Establish standards, documentation, and debugging workflows that allow cross-functional teammates to diagnose issues and extend the system with minimal friction.
• Contribute to a culture of reliability and performance by proactively identifying bottlenecks, proposing architectural improvements, and quantifying their impact.
• Act as a technical owner for the training infrastructure, responding to incidents, prioritizing roadmap work, and mentoring other engineers on best practices.

Requirements

• Demonstrate proficiency in software engineering with production experience using PyTorch and Python at scale.
• Bring a proven track record of building or scaling training infrastructure for large-scale machine learning, robotics, or control systems.
• Show familiarity with physics simulation platforms such as PyBullet, NVIDIA PhysX, Warp, or MuJoCo in real-world projects.
• Exhibit a solid understanding of robotics systems, dynamics, and control theory relevant to humanoid manipulation and locomotion.
• Have direct experience with imitation learning, reinforcement learning, or policy distillation methods applied to challenging control problems.
• Possess the capability to take full ownership of systems, including designing, debugging, and optimizing components that support colleagues across the organization.
• Provide evidence of modeling photorealistic simulation environments and contact interactions for complex manipulation in prior work.
• Highlight experience ensuring that simulation fidelity aligns with the practical constraints of hardware testing and safety validation.

Nice to have

• Hands-on experience with legged or humanoid robot control in research or industry settings.
• Background in cluster management, job schedulers, or distributed systems, including container orchestration and resource isolation.
• Prior experience moving control policies or machine learning models from simulation onto real-world humanoid hardware.

Practical notes

The annual base salary for this role is $150,000 to $300,000. Final compensation is determined by individual skills, experience, and job-related knowledge. Additional benefits and compensation components may be provided and will be discussed upon the extension of an offer. The role is based in San Jose, Canada, and requires full-time, five-day presence in the office.