Embodied AI / VLA Research Engineer

International Recruiting LLC•San Jose, CA

About The Position

Research, train, optimize, and deploy embodied AI models, including Vision-Language-Action (VLA) models, World Models, and related robotics foundation models. The role spans multimodal perception, robot control, long-horizon task planning, large-scale robot data, and model deployment, combining model development with real-world robotic system integration. The engineer develops models for robotic manipulation, multimodal perception and control, VLA reasoning, long-horizon planning, and robot-environment interaction, translating research ideas into practical systems that operate reliably on physical robots. The role also involves training and evaluating models on large-scale real-robot and egocentric human-demonstration datasets; designing multimodal approaches using vision, force, tactile sensing, proprioception, and other signals; and building scalable robotics data pipelines. The position includes deploying trained models to physical robots, debugging and testing real-world systems, diagnosing issues across models, sensors, software, and hardware, and iterating between model training and robot testing to improve performance and reliability. The engineer works with data and robotics teams and takes technical problems from research through implementation in a fast-paced startup environment.

Requirements

At least 1 year of relevant industry or research experience in machine learning, robotics, computer vision, embodied AI, or a related field.
Strong understanding of deep learning and modern machine learning methods.
Experience with PyTorch or similar deep learning frameworks.
Experience training and evaluating machine learning models.
Strong Python programming skills and familiarity with relevant ML or robotics tooling.
Understanding of multimodal learning, computer vision, robotics, or related areas.
Ability to work in a fast-paced startup environment and take ownership of technical problems from research through implementation.
Strong problem-solving and debugging skills.

Nice To Haves

Experience deploying and debugging machine learning models on physical robot systems.
Ability to bring models from development into real-world robotic environments.
Experience troubleshooting and stabilizing robotic systems in production or experimental environments.
Experience working with force, tactile, or other multimodal sensing.
Experience designing or training multimodal fusion models.
Hands-on experience with VLA models, including model design, training, evaluation, or real-world applications.
Experience with robotic manipulation or embodied AI systems.
Experience with large-scale distributed model training.
Familiarity with DDP, DeepSpeed, FSDP, or similar distributed training frameworks.
Experience optimizing training performance, GPU utilization, memory usage, or training throughput.
Experience working with large-scale datasets and distributed data pipelines.

Responsibilities

Research and develop VLA, World Model, and other embodied AI foundation models.
Develop models for robotic manipulation, multimodal perception and control, long-horizon task planning, VLA reasoning, and robot-environment interaction.
Optimize models for real-world robotic applications and translate research ideas into practical, reliable systems for physical robots.
Train and optimize embodied AI foundation models using large-scale real-robot datasets and egocentric human-demonstration data.
Design approaches for incorporating vision, force, tactile sensing, proprioception, and other robot or environmental signals.
Develop and evaluate multimodal fusion architectures.
Conduct model training, evaluation, benchmarking, and performance optimization.
Analyze model performance and identify improvements to training efficiency, generalization, and real-world performance.
Build and improve large-scale robotics data pipelines, including processes for cleaning, filtering, resampling, augmentation, and data-quality evaluation.
Work with data and robotics teams to improve dataset quality and training efficiency.
Deploy trained models to physical robot systems and conduct real-world debugging, testing, and optimization.
Diagnose issues involving models, sensors, software, and robotic hardware.
Iterate between model training and real-world testing to improve system performance and reliability.