Machine Learning Engineer, VLA & RL - Senior

Full-timeTC starting at 70k

Milan / Zurich

Posted 1 day ago

Build vision-language-action and reinforcement learning models for real-world robotic systems. Train policies that generalize across embodiments, tasks, simulators, and physical deployments.

Cyberwave is building the infrastructure layer for intelligent machines - making robotics as accessible, scalable, and programmable as cloud software. Our platform connects simulation, digital twins, edge devices, cloud training, and real robots into one operating layer for robotics teams.

We're looking for a Machine Learning Engineer focused on vision-language-action (VLA) models, reinforcement learning, and cross-embodiment transfer. You'll work on models that turn perception, language, and task context into robot actions across different hardware platforms: arms, mobile robots, drones, and other industrial systems.

This is a hands-on applied ML role. We care about candidates who have trained and evaluated real policies, debugged failures across simulation and hardware, and understand the gap between promising demos and reliable deployment. You'll work closely with robotics, simulation, infrastructure, and product teams to build learning systems that can be trained at scale, evaluated rigorously, and deployed safely on real robots.

This role is based in Milan or Zurich, with regular access to real robots, simulation infrastructure, and customer-facing deployment scenarios.

Vision-Language-Action ModelsReinforcement LearningCross-Embodiment TransferImitation LearningBehavior CloningPyTorchPythonRobotics SimulationSim2real TransferPolicy EvaluationEmbodied AIRobot Control

Work Style

Hands-on applied ML for embodied AI, simulation, and real robot deployments

Requirements

• 3+ years of hands-on experience building ML systems for robotics, embodied AI, reinforcement learning, or visuomotor control
• Specific experience with vision-language-action (VLA) models, robotic foundation models, imitation learning, behavior cloning, or language-conditioned policies
• Strong experience with reinforcement learning algorithms and workflows, such as PPO, SAC, offline RL, RL fine-tuning, reward modeling, or policy evaluation
• Practical experience with cross-embodiment transfer, including transferring policies across robot morphologies, sensors, action spaces, simulators, or real hardware platforms
• Experience training and evaluating policies in simulation environments such as MuJoCo, Isaac Sim/Lab, PyBullet, ManiSkill, robosuite, or similar robotics simulators
• Strong Python and PyTorch skills, with good software engineering habits for reproducible training, experiment tracking, datasets, and evaluation
• Comfort debugging model failures across perception, action representations, control loops, latency, data quality, and hardware behavior
• Comfortable working in English in an international, fast-moving environment
  
  

Responsibilities

• Train and evaluate VLA, imitation learning, and reinforcement learning policies for real robotic tasks
• Build model and data pipelines for language-conditioned robot control, visuomotor policies, trajectory datasets, and action representations
• Design experiments for cross-embodiment transfer across arms, mobile robots, drones, simulated systems, and physical hardware
• Improve policy robustness through simulation, domain randomization, dataset curation, offline evaluation, online rollouts, and sim-to-real validation
• Collaborate with robotics and infrastructure teams to deploy learned policies into Cyberwave's edge, simulation, and digital twin stack
• Create rigorous evaluation suites for task success, generalization, safety, latency, and real-world reliability
• Stay close to frontier research in embodied AI while turning useful ideas into production-quality systems
  
  

What We Offer

Work on frontier embodied AI with direct paths to real robot deployment

Access to real robots, simulation infrastructure, and robotics datasets

Competitive compensation and meaningful equity

Join a high-talent team of repeat founders, ex-Google engineers, and PhDs

In-person collaboration in Milan and Zurich with real hardware and high ownership