Robo Engineer

Lightwheel; San Francisco (CA); Vor Ort; USD 150.000 - 210.000; Vollzeit

Zusammenfassung

Lightwheel in San Francisco is seeking a role focused on building a robust data-model-evaluation closed loop and underlying infrastructure for data, evaluation and training. You will contribute to experiment design and algorithm research across embodied AI, including data collection, simulation data, and model training pipelines, with emphasis on data quality and efficient feedback loops.

The position requires hands-on robotics data, reinforcement learning expertise, and the ability to

Qualifikationen

• Experience with robotics data collection and dataset curation.
• Strong RL knowledge; familiarity with World Models and simulation.
• Ability to design experiments and evaluate data-model-evaluation loops.

Aufgaben

• Develop data and evaluation pipelines within the data-model-evaluation loop.
• Contribute to model architectures and training pipelines for embodied AI.
• Collaborate with researchers to design experiments and analyze results.

Kenntnisse

Robotics data

Reinforcement learning

World Models

Simulation

Tools

Claude Code

Jobbeschreibung

This role focuses on building a robust data-model-evaluation closed loop, developing the underlying infrastructure for data, evaluation and training within this loop, and contributing to experiment design and algorithm research. Key algorithm and development work spans the following research areas:

• For embodied AI datasets, research data collection, data distribution construction and data proportioning strategies; build a data pyramid consisting of human-collected data, simulation data and real-robot data.
• Analyze the distribution and quality of diverse robot and human datasets, build systems for data distribution and quality management, and investigate how data distribution and quality impact downstream model performance.
• Research technical directions for embodied AI models including VLA, World Model, Offline/Online RL and In-context Learning, and design model architectures and training pipelines.
• Conduct end-to-end optimization from the perspective of the data-model-evaluation loop. Explore the development of an efficient learning system, design real-robot and simulation evaluation pipelines to generate feedback, and build an automated, rapidly iterating data flywheel.

Requirements

• Solid hands-on experience in robotics data, World Models, reinforcement learning and simulation, with a sound understanding of underlying algorithm principles.
• Strong engineering capabilities. Able to independently debug robot systems, cloud data processing pipelines, model training workflows and robot rollout setups.
• Self-driven problem-solving skills. Comfortable leveraging coding assistant tools such as Claude Code; sufficient token quota will be provided for this role.
• Basic English communication skills to collaborate with international researchers.