Forward Deployed Engineer

Description

The role connects the company's robotics foundation model data engine with Embodied AI and World Model R&D teams around the world, and is titled Forward Deployed Engineer (FDE - Robotics Data Direction).

Most FDE roles centre on on-site hardware installation and driver debugging. This one does not, because the company's platform already handles data streaming and simulation at a high level.

The goal is to understand what clients need from academic and commercial projects, obtain and match dataset specs, build high-quality multimodal motion and scene simulation data, enforce strict data governance, and run fast, repeated checks inside and outside the team. The data you deliver should help client robots generalise better.

This is a startup environment with rapid iteration and requirements that are still taking shape.

Responsibilities

• Talk directly with Embodied AI and World Model researchers to establish what their models need as input, such as camera field of view, LiDAR precision, arm dynamic limits and trajectory action formats.
• Turn vague business and research problems into precise Dataset Specs that define data dimensions, sensor parameters, control action spaces and metadata formats.
• Plan target datasets that mix routine scenarios with long-tail corner cases, so they generalise well and cover out-of-distribution extreme conditions.
• Review client historical or public datasets to find gaps in geography, lighting, action types and obstacle layout, then fill those gaps with targeted data.
• Use and extend the company's 3D simulators, such as Isaac Sim and MuJoCo, or generative world model tools to run large-scale synthesis pipelines.
• Write Python or Bash scripts that generate custom robot trajectories, multi-angle video streams, 3D point clouds and dynamics states.
• Curate, filter and calibrate data, removing collision errors, brief sensor dropouts and action drift.
• Write automated QA scripts for static and dynamic checks across tens of millions of frames, covering dynamics reachability, temporal alignment and label accuracy.
• Oversee final packaging into MCAP, HDF5 or LeRobot formats so the data loads directly into training pipelines.
• Own each delivery, then track how well the data performs during the client's early model training, feeding results back into a data-to-model performance loop.
• Speak with the Algorithm R&D and Platform Development teams every day, passing system bugs and shared needs found in the field so the core data engine is upgraded in a standard way.

Requirements

• Write Python well and follow solid software engineering habits, including Git, Docker and Shell.
• Have worked with large multimodal data tools such as Pandas, NumPy, Arrow, HDF5, and RLDS or TensorFlow Datasets structures.
• Have a firm grasp of 3D spatial maths: quaternions, rotation matrices, coordinate transforms, forward and inverse kinematics of robot arms, and sensor intrinsic and extrinsic parameters.
• Communicate clearly across teams, aligning technical limits and Specs with top AI researchers and with non-technical business staff.
• Work independently with a results focus, pushing projects forward through rapid iteration and unclear requirements.
• Preferred: hands-on project experience with physics simulators such as Isaac Sim, MuJoCo, PyBullet, Unity or Unreal Engine, or with NeRF, 3DGS or generative world models. Highly preferred.
• Preferred: understanding of autonomous driving perception, planning and control, or of training and evaluating Embodied AI foundation models such as VLA or RT-2. Highly preferred.
• Preferred: fluent in Mandarin Chinese and English, with strong written and spoken skills in both, able to convey technical requirements, updates, Specs and operational issues across US and China-based teams.

Conditions

The work is 100% data- and software-driven, with no on-site physical hardware deployment or mechanical debugging.

Benefits

• Default benefits: health insurance, vision care, dental coverage, 401(k), paid holidays, PTO and sick leave.