Embodied AI / VLA Research Engineer

Description

You will research, train, tune and deploy embodied AI models, including vision-language-action (VLA) models, world models and other robotics foundation models.

The work covers multimodal perception, robot control, long-horizon planning, large robot datasets and model deployment, with hands-on time on both model building and integration into real robots.

Responsibilities

• Research and build VLA, world model and other embodied AI foundation models.
• Focus areas include robotic manipulation, multimodal perception and control, long-horizon planning, reasoning that links vision, language and action, and how robots interact with their surroundings.
• Turn research ideas into models and systems that run reliably on physical robots.
• Develop and optimize models for real robot use cases.
• Train and tune foundation models on large real-robot datasets and egocentric human demonstration data.
• Build ways to feed in vision, force, tactile, proprioceptive and other robot and environment signals.
• Design and test multimodal fusion architectures.
• Run training, evaluation and benchmarking, and improve training speed and performance.
• Study results to find ways to boost training efficiency, generalization and real-world performance.
• Build and extend large-scale robotics data pipelines for training.
• Own processes for cleaning, filtering, resampling, augmenting and quality-checking data.
• Design pipelines that scale to large volumes of robot and human demonstration data.
• Partner with data and robotics teams to raise dataset quality and training efficiency.
• Deploy trained models onto physical robots.
• Debug, test and tune performance on real robots.
• Find root causes across models, sensors, software and robot hardware.
• Cycle between training and real-world testing to improve the system.
• Help keep models stable and consistent in real environments.

Requirements

• At least 1 year of relevant industry or research experience in machine learning, robotics, computer vision, embodied AI or a related field.
• Solid grasp of deep learning and current machine learning methods.
• Hands-on use of PyTorch or a similar deep learning framework.
• Experience training and evaluating machine learning models.
• Strong Python skills and comfort with common ML and robotics tools.
• Understanding of multimodal learning, computer vision, robotics or adjacent areas.
• Able to take technical problems from research through to implementation in a fast-paced startup.
• Strong problem-solving and debugging ability.
• Strong plus: experience deploying and debugging machine learning models on physical robots.
• Strong plus: experience bringing models from development into real robot environments.
• Strong plus: experience troubleshooting and stabilizing robot systems in production or experimental settings.
• Strong plus: experience with force, tactile or other multimodal sensing.
• Strong plus: experience designing or training multimodal fusion models.
• Strong plus: hands-on work with VLA models, covering design, training, evaluation or real-world use.
• Strong plus: experience with robotic manipulation or embodied AI systems.
• Strong plus: experience with large-scale distributed model training.
• Strong plus: familiarity with DDP, DeepSpeed, FSDP or similar distributed training frameworks.
• Strong plus: experience optimizing training speed, GPU use, memory or throughput.
• Strong plus: experience with large datasets and distributed data pipelines.
• Ideal candidate: excited by where foundation models meet physical robotics.
• Ideal candidate: hands-on, and comfortable moving between research, coding, experiments and robot testing.
• Ideal candidate: interested in problems that simulation or software alone cannot solve.
• Ideal candidate: comfortable with large datasets and modern foundation model architectures.
• Ideal candidate: owns a problem end to end, from data to training, evaluation, deployment and real-world iteration.
• Ideal candidate: fine with an early-stage team where priorities shift quickly.
• Ideal candidate: keeps up with new VLA, world model and embodied AI research and turns it into working systems.

Benefits

• Work on embodied AI and robotics foundation models directly.
• Use large real-world robot and human demonstration datasets.
• Get hands-on exposure to the full AI lifecycle, from data pipelines to deployment on real robots.
• Join a fast-moving startup with substantial ownership and technical independence.
• Work alongside teams at the leading edge of robotics and foundation model development.
• Default benefits: health insurance, vision care, dental coverage, 401(k), paid holidays, PTO and sick leave.