The company builds automated labs that combine AI, robotics, and life sciences to design, run, and learn from experiments. It does not develop drugs itself. It builds the infrastructure for scientists who do, with the goal of speeding up drug discovery and building that capacity in the United States.
Chemistry work demands precise, dexterous handling of instruments, materials, and consumables built for human hands. Autonomous labs therefore need robots that can learn the wide range of routine tasks scientists perform, adjust when conditions shift, and work dependably in the physical world.
The team is hiring a Member of Technical Staff for Robot Learning to build the learning systems behind this. You will own the path from expert demonstrations and robot experience through policy training, evaluation, and deployment, working across robot learning, data, controls, hardware, and chemistry to make capable models perform difficult scientific tasks reliably.
Responsibilities
Develop mid-training and post-training methods for robot policies using real-world demonstrations, robot experience, and task-specific data.
Build systems that capture expert chemists' demonstrations of difficult procedures and turn them into behaviors robots can repeat and apply to new cases.
Develop and adapt world models, world-action models, vision-language-action models, and other learned representations for precise, contact-rich manipulation.
Build training and evaluation systems that ingest, process, curate, and learn from large multimodal robot datasets.
Deploy policies on physical robots and find the cause of failures across models, data, perception, controls, hardware, and task design.
Run the full cycle of data collection, training, evaluation, and deployment. Decide which data or training change is needed, check whether it helps, and use each deployment to shape the next iteration.
Work closely with chemists, hardware engineers, infrastructure engineers, and robot operators to design tasks, gather useful experience, and move research results into reliable lab capabilities.
Requirements
We weigh what you have built more heavily than your job title or years of experience.
Solid grounding in machine learning, deep learning, and robot learning.
Experience with post-training, imitation learning, reinforcement learning, behavior cloning, or fine-tuning learned policies.
Experience training, adapting, and debugging modern deep learning models in PyTorch or JAX.
A record of designing rigorous experiments that produced measurable gains in model or policy behavior.
A strong sense of data quality and makeup: which experience a model needs, where its data falls short, and how to fix that efficiently.
Experience building data pipelines, training systems, or evaluation infrastructure for multimodal or robotics data.
Experience with real physical robots, or a strong record building learned systems that operate in complex physical settings.
The ability to trace failures across the whole system instead of treating model, data, controls, and hardware as separate problems.
High ownership, and enjoyment of problems where neither the requirements nor the right technical approach are clear yet.
Helpful but not required:
Nice to have:
World models, world-action models, or vision-language-action models
Post-training or adapting large pretrained models for robotics
Large-scale multimodal datasets, dataset curation, and distributed training
Imitation learning, behavior cloning, reinforcement learning, or offline RL
Learning from human demonstrations, teleoperation data, or real-world robot experience
Diffusion policies, action chunking, or other modern visuomotor policy architectures
Dexterous, precise, or contact-rich manipulation
Multimodal representations spanning vision, language, actions, sensors, and robot state
Simulation, sim-to-real transfer, or synthetic robot data
Real-world robot evaluation, benchmarking, and failure analysis
Example problems you might work on:
Taking a capable pretrained model and post-training it on a new chemistry task using demonstrations from expert chemists.
Ingesting, curating, and training on millions of trajectories that combine video, actions, robot state, language, and other sensor data efficiently.
Deciding what data to collect when a robot completes 80% of a task but consistently fails at one particular stage.
Working out whether a failure came from the policy, training data, perception, controls, hardware, or how the task was specified.
Adapting policies to precise, contact-rich work with powders, liquids, vials, caps, pipettes, and instruments designed for human hands.
Judging whether a policy is truly getting more capable instead of overfitting to one setup, object, or demonstration.
Success in this role:
You set up a repeatable cycle running from expert demonstrations and robot experience through training, evaluation, and deployment. Policies get measurably more reliable on real chemistry tasks, and failures drive clear choices about what data to gather, what experiments to run, and which parts of the system to improve.
The robots learn new tasks faster, recover from a wider set of failures, and generalize across materials, equipment, and lab conditions. Each round of real-world experience makes the learning system, and the autonomous lab built on it, more capable.