Robotics Research Engineer

Robotics Research Engineer
Компания: Zeromatter
Город: Mountain View
As a Robotics Research Engineer on our Robotics team, you'll design, train, and evaluate robotics policies from imitation learning to reinforcement learning to foundation/VLA models using Zeromatter's simulation platform, then validate, and iterate against real hardware. This work is grounded in real customer use cases as much as internal research goals. The manipulation and automation tasks you train and validate often trace directly back to a problem a customer is trying to solve. You'll work closely with our Dynamics team, who own Physics Engine development, to make sure the simulator keeps pace with what policy training needs. You'll be running the experiments, in sim and in the lab, that push our simulator's fidelity and our policies' transfer performance forward together. Responsibilities may vary from month to month, but here are some examples of things you might find yourself working on. Responsibilities ---------------- * Design and run experiments training robotics policies (imitation learning, reinforcement learning, VLA models, etc.) entirely or primarily in simulation. * Work hands-on with physical robots (e.g. bimanual manipulator arms) to collect real-world data, validate sim-trained policies, and characterize the sim-to-real gap. * Collaborate with our Dynamics team to identify simulation fidelity gaps that limit transfer performance, and help prioritize and validate fixes. * Build and maintain data pipelines and evaluation frameworks/benchmarks for policy training and transfer performance. * Prototype new tasks and domains (e.g. dexterous manipulation, cross-embodiment transfer), often grounded in real customer use cases, to expand what Zeromatter's platform can demonstrate. * Read, evaluate, and adapt techniques from state-of-the-art robot learning research for use on Zeromatter's platform. * Contribute findings and results to customer-facing technical materials and demonstrations. * Build reusable libraries and tooling (in Python, Rust, and/or C++) to support this work at scale. Qualifications -------------- * You have an engineering or related technical degree (e.g. robotics, computer science, mechanical/electrical engineering, physics) or equivalent hands-on experience. * Experience training and evaluating robot learning policies (imitation learning, reinforcement learning, and/or foundation/VLA models) gained in industry, in a research lab, or through graduate/postdoctoral work. * Hands-on experience working with physical robots including collecting data, running experiments, and debugging real-world hardware/software issues. * Proficient in Python, Rust, and/or C++, with experience in ML frameworks such as PyTorch, TensorFlow, or JAX. * Comfortable working with incomplete information, iterating quickly between simulation and real-world validation. * Familiarity with the practical challenges of sim-to-real transfer (e.g. domain randomization, system identification, observation/action space design, adaptation). * Bonus: Experience with dexterous manipulation or other contact-rich manipulation tasks. * Bonus: Experience with cross-embodiment policy learning or transfer across robot platforms. * Bonus: Experience with teleoperation systems and data collection pipelines. * Bonus: Experience working end-to-end across the robotics stack, including perception, classical and learned control systems, task planning, etc. * Bonus: Familiarity with physics/dynamics simulation engines (e.g. MuJoCo, Isaac Sim, or custom/proprietary engines). * Bonus: Publication record or open-source contributions in robot learning, embodied AI, or manipulation.