Job description

Innodata (Nasdaq: INOD) is a global data engineering company. We believe that data and Artificial Intelligence (AI) are inextricably linked. Our mission is to enable the responsible advancement of artificial intelligence by providing the data, evaluation frameworks, and human expertise required to build AI systems that can be trusted at scale. We provide a range of transferable solutions, platforms, and services for Generative AI / AI builders and adopters. In every relationship, we honor our 36+ year legacy delivering the highest quality data and outstanding outcomes for our customers. Scope of the Role: Progress in physical AI is gated as much by data design as by model architecture. The best manipulation and locomotion policies, vision-language-action models, and world models are only as good as the trajectories, demonstrations, and environments they learn from — so the highest-leverage decisions are made long before a model trains: what to capture, what to synthesize, how to represent an action, and how to know whether any of it will transfer to a real robot. Innodata is building the data and evaluation practice behind the next generation of robotics foundation models, in direct partnership with the customers and frontier labs defining that frontier. We are hiring a Research Scientist to own the science of that data. You will partner directly with the customers and frontier labs building robot foundation models, as interested in the data those models learn from as in the models themselves. Your leverage is scientific judgment about data: deciding what is worth capturing in the real world versus generating in simulation, how to weight a training mix across embodiments and sensors, and how to measure whether a policy will hold up outside the lab. Your conclusions will shape what our partners collect next. What You’ll Own: You will define how Innodata designs, structures, and evaluates data for robot foundation models, and you will validate those choices experimentally. Concretely, you will: Translate the requirements of robotics foundation models — vision-language-action models, world models, and manipulation and locomotion policies — into concrete data specifications: modalities, action representations (tokenization, chunking, diffusion and flow-matching action experts), sampling, annotation schemas, and evaluation criteria. Decide what is worth capturing in the real world versus generating in simulation, and curate and weight training mixes across heterogeneous robot datasets that span different embodiments, action spaces, and sensor setups. Guide collection across capture modalities — motion capture, egocentric, exocentric, teleoperation, and multi-sensor — and across synthetic pipelines, so that what we produce maps cleanly to model needs rather than accumulating for its own sake. Build evaluation and benchmarking methodology that predicts real-world transfer — coverage, discriminative power, reliability, and sim-to-real fidelity — including world-model evaluation (rollout quality and action-conditioned prediction) and the domain-randomization and system-identification choices that close the sim-to-real gap. Run the experiments that prove it: fine-tune and evaluate foundation models on Innodata data, with data-quality ablations and scaling studies that demonstrate specific data decisions produce measurable model improvement. Design adversarial and long-horizon evaluations that surface where policies and world models break, and turn those failure modes into better data. Publish. Turn what you learn into benchmarks, methodology, and papers that advance the field and earn the trust of the customers and frontier labs we partner with. Collaborate with the capture lab, annotation teams, and the synthetic-data pipeline to turn specifications into operational collection and labeling plans. You’ll Thrive in This Role If You Have: Roughly 4+ years of hands-on industry experience in robot learning or robotics ML. We weight practical experience over formal credentials; because the field is young, a PhD with a compelling, current research agenda can offset the lower end, and we will also consider very senior candidates with the right background. A Bachelor's degree in computer science, electrical engineering, robotics, or a related technical field is required; an advanced degree (MS or PhD) in a relevant field is preferred. Trained and evaluated robot policies yourself — manipulation or locomotion via imitation learning or reinforcement learning — with strong PyTorch fundamentals. You build and measure models, you don't just call them. A way of thinking in datasets: you have curated, filtered, and weighted robot data across embodiments and sensors, and you have real opinions about what makes it good for a given objective. Fluency in robotics data formats and standards — the LeRobot dataset format, RLDS, and Open X-Embodiment — along with common motion and sensor formats. Hands-on experience with simulation and synthetic data — NVIDIA Isaac Sim, Isaac Lab, and Omniverse, or MuJoCo and comparable engines — including domain randomization, system identification, and sim-to-real transfer. Experience with teleoperation or egocentric data collection, and comfort adapting VLM backbones for control and fine-tuning large VLA models with the modern toolchain (HuggingFace transformers, PEFT). A track record the field recognizes: first-author publications or strong open-source contributions at venues such as CoRL, ICRA, IROS, RSS, or NeurIPS. The ability to work directly with the research scientists at the customers and frontier labs we partner with, and to explain data and modeling decisions clearly to both expert and non-expert audiences, backed by a rigorous, reproducible approach to experiments and documentation. Bonus: interest or hands-on experience in responsible-AI evaluation and red-teaming for embodied systems, for example safety and robustness testing. The expected salary range for this position is $160,000 - $185,000 p/year, based on experience, skills, and qualifications. Please be aware of recruitment scams involving individuals or organizations falsely claiming to represent employers. Innodata will never ask for payment, banking details, or sensitive personal information during the application process. To learn more on how to recognize job scams, please visit the Federal Trade Commission’s guide at consumer.ftc.gov. If you believe you’ve been targeted by a recruitment scam, please report it to Innodata at verifyjoboffer@innodata.com and consider reporting it to the FTC at ReportFraud.ftc.gov .

Requirements

Must-haves

You will define how Innodata designs, structures, and evaluates data for robot foundation models, and you will validate those choices experimentally. Concretely, you will:
Run the experiments that prove it: fine-tune and evaluate foundation models on Innodata data, with data-quality ablations and scaling studies that demonstrate specific data decisions produce measurable model improvement.
Design adversarial and long-horizon evaluations that surface where policies and world models break, and turn those failure modes into better data.
Publish. Turn what you learn into benchmarks, methodology, and papers that advance the field and earn the trust of the customers and frontier labs we partner with.
Collaborate with the capture lab, annotation teams, and the synthetic-data pipeline to turn specifications into operational collection and labeling plans.
You’ll Thrive in This Role If You Have:
Trained and evaluated robot policies yourself — manipulation or locomotion via imitation learning or reinforcement learning — with strong PyTorch fundamentals. You build and measure models, you don't just call them.
A way of thinking in datasets: you have curated, filtered, and weighted robot data across embodiments and sensors, and you have real opinions about what makes it good for a given objective.
Fluency in robotics data formats and standards — the LeRobot dataset format, RLDS, and Open X-Embodiment — along with common motion and sensor formats.
Hands-on experience with simulation and synthetic data — NVIDIA Isaac Sim, Isaac Lab, and Omniverse, or MuJoCo and comparable engines — including domain randomization, system identification, and sim-to-real transfer.
Experience with teleoperation or egocentric data collection, and comfort adapting VLM backbones for control and fine-tuning large VLA models with the modern toolchain (HuggingFace transformers, PEFT).
A track record the field recognizes: first-author publications or strong open-source contributions at venues such as CoRL, ICRA, IROS, RSS, or NeurIPS.
If you believe you’ve been targeted by a recruitment scam, please report it to Innodata at verifyjoboffer@innodata.com and consider reporting it to the FTC at ReportFraud.ftc.gov .

Nice-to-haves

A Bachelor's degree in computer science, electrical engineering, robotics, or a related technical field is required; an advanced degree (MS or PhD) in a relevant field is preferred.
Bonus: interest or hands-on experience in responsible-AI evaluation and red-teaming for embodied systems, for example safety and robustness testing.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aiml

Platform: greenhouse
Effort: quick

Benefits

Benefits detected in the listing.

Vision