Research Scientist, Robotics – World Models

This is a fully remote position, open to applicants in United States.

📋 Description

• Define the methods by which Innodata designs, structures, and evaluates data for foundational models in robotics.

• Convert requirements for robotics foundation models into comprehensive data specifications, encompassing modalities, action representations, sampling, annotation schemas, and evaluation criteria.

• Determine what aspects should be captured in the real world as opposed to those that can be generated in a simulated environment.

• Curate and prioritize training mixes across diverse robot datasets, embodiments, action spaces, and sensor configurations.

• Direct data collection efforts across various methodologies including motion-capture, egocentric, exocentric, teleoperation, multi-sensor, and synthetic data pipelines.

• Develop evaluation and benchmarking techniques that forecast real-world transfer effectiveness, including world-model assessments and sim-to-real methodologies.

• Fine-tune and assess foundation models utilizing Innodata's data through processes of data-quality ablation and scaling analysis.

• Design adversarial and long-term evaluations, transforming failure modes into enhanced data.

• Publish benchmarks, methodologies, and research papers.

• Collaborate with teams focused on data capture, annotation, and synthetic data to implement collection and labeling strategies.

⛳️ Requirements

• Approximately 4+ years of practical experience in robot learning or robotics machine learning within the industry.

• A Bachelor's degree in computer science, electrical engineering, robotics, or a related technical discipline is mandatory.

• Experience training and evaluating robot policies through imitation learning or reinforcement learning techniques.

• Strong foundational knowledge of PyTorch.

• Proficiency in curating, filtering, and weighting robot data across various embodiments and sensor types.

• Familiarity with the LeRobot dataset format, RLDS, Open X-Embodiment, and standard motion and sensor formats.

• Practical experience with NVIDIA Isaac Sim, Isaac Lab, Omniverse, MuJoCo, or similar simulation platforms.

• Knowledge in domain randomization, system identification, and sim-to-real transfer techniques.

• Experience with teleoperation or egocentric data collection methodologies.

• Aptitude for adapting VLM backbones for control and fine-tuning large VLA models utilizing HuggingFace transformers and PEFT.

• First-author publications or significant contributions to open-source projects at respected venues such as CoRL, ICRA, IROS, RSS, or NeurIPS.

• Capability to collaborate effectively with customers and research scientists in frontier labs, articulating data and modeling choices with clarity.

• A rigorous and reproducible approach to conducting experiments and maintaining documentation.

🏝️ Benefits

• Competitive salary and performance-based bonuses.

• Comprehensive health, dental, and vision insurance.

• Opportunities for professional development and continuous learning.

• Flexible working hours and the possibility for remote work.

• Engaging and inclusive company culture.