VLA/WAM Integration & Deployment Engineer

Flexion Robotics • San Francisco, CA

2d • Onsite

About The Position

At Flexion, the team is building an intelligence layer for the next generation of humanoid robots and aims to accelerate the transition from fragile prototypes to real-world deployment. The company was founded by scientists in robot reinforcement learning and is backed by international venture capital firms.

The VLA/WAM Integration & Deployment Engineer turns VLA and World-Action Model research into reliable capabilities running on real humanoid robots. The role spans the deployment loop: integrating foundation models with agentic systems, optimizing inference, collecting targeted real-world data, conducting rigorous evaluations, and using deployment experience to improve training. It is a broad, hands-on position focused on making advanced models work as part of a complete autonomous system.

The posting lists locations in Zurich and San Francisco. It describes a central Zürich office with robotics testing facilities and infrastructure, and provides location-specific benefits for both locations.

Requirements

• Strong Python and PyTorch skills and experience working with modern deep-learning systems.
• Experience deploying learning-based models in robotics, autonomous systems, or another real-time embodied setting.
• Familiarity with VLA, VLM, diffusion, flow-matching, video/world-model, or related multimodal architectures.
• Strong systems intuition around inference latency, model serving, data pipelines, logging, and debugging complex ML systems.
• Experience designing rigorous evaluation and diagnosing failures in learned systems.
• Comfort working directly with robots, sensors, data collection, and imperfect real-world hardware.
• Ability to work across model training, deployment, data, infrastructure, and integration rather than within a narrowly defined research area.

Nice To Haves

• Experience with agentic AI systems, tool use, hierarchical policies, planning, or closed-loop verification.
• Experience with TensorRT, torch.compile, quantization, model compression, or other inference optimization techniques.
• Experience with teleoperation, human-in-the-loop learning, autonomous data collection, or online/post-training workflows.
• Experience with humanoids, dexterous manipulation, loco-manipulation, or whole-body control.
• Familiarity with reinforcement learning, continual learning, and sim-to-real.

Responsibilities

• Deploy and validate learned policies across multiple humanoid platforms and real-world tasks.
• Optimize model serving, latency, action generation, asynchronous execution, and closed-loop control for real-time robot operation.
• Integrate VLA/WAM policies with higher-level agents for task decomposition, policy selection, structured conditioning, verification, recovery, and replanning.
• Develop improved workflows for teleoperation, interventions, autonomous rollouts, failures, recoveries, and targeted post-training data collection.
• Develop reproducible real-robot evaluations, regression suites, failure analysis, and principled comparisons between model versions.
• Convert real-world successes and failures into datasets and experiments that improve subsequent pre-training and post-training.

Benefits

• Competitive compensation.
• Enhanced pension plan.
• Enhanced holiday and paid leave perks.
• Relocation and permit sponsorship.
• Central Zürich office with top-tier robotics testing facilities and infrastructure.
• 401(k) with company contributions (San Francisco).
• Health, dental, and vision coverage with the flexibility to choose your own plan (San Francisco).
• Open PTO policy and paid company holidays (San Francisco).

What This Job Offers

• Job Type: Full-time
• Education Level: No Education Listed