Atlas VLA Research Intern

Boston Dynamics; Waltham (MA); On-site; USD 41,000 - 96,000; Full time

Job summary

Boston Dynamics is seeking PhD-level interns to join the Atlas VLA Research Team and advance vision-language-action models on real-world robot hardware. You will lead experiments, collect data, and prototype new training recipes that bridge simulation and Atlas deployment.

You will collaborate with engineers and scientists to integrate learned policies with robot control, extend data pipelines, and contribute production-grade code in Python and C++.

Qualifications

• Actively pursuing a PhD (preferred) or a research-heavy Master’s in CS/Robotics.
• Experience training and deploying imitation learning models for robotics with strong experimental hygiene.
• Familiarity with deploying ML on real hardware and robotics platforms.

Responsibilities

• Lead high-stakes research projects establishing hypotheses and experiments for VLA/WAM performance.
• Prototype and deploy data collection/mixing/grading schemes and training recipes on Atlas hardware.
• Collaborate across teams to integrate learned policies with low-level robot control.
• Extend data pipelines to ingest diverse data sources for robustness.
• Write production-grade Python/C++ code integrating with existing systems.

Skills

Python

C++

Imitation learning

Research

Education

PhD in CS/Robotics

Tools

Kubernetes

SLURM

Open X-Embodiment

Job description

Are you passionate about the intersection of foundation models and physical AI? As an intern on the Atlas VLA Research Team , you will join a world-class group of engineers and scientists dedicated to giving Atlas the intelligence to perceive, reason, and act.

Our team focuses on scaling Vision-Language-Action (VLA) models that leverage diverse data sources and training recipes to keep Atlas grounded and capable of performing challenging dexterous tasks. We are looking for PhD-level interns to lead research projects that push the boundaries of what humanoid robots can do. You won’t work on pure theory; you’ll be deploying your ideas directly onto the most sophisticated pieces of hardware on the planet.

How you will make an impact:

• Lead a high-stakes research project establishing hypotheses and experiments for VLA / WAM performance, leveraging and extending our benchmarking, training, and evaluation with various data sources (teleoperation, UMI-style data, VQA, etc.).
• Prototype and deploy your data collection / mixing / grading schemes, pre- / mid- / post-training recipes, and algorithms directly on Atlas, moving from simulation to hardware.
• Collaborate across teams to integrate learned policies with low-level robot control.
• Extend data pipelines that ingest alternative data sources to improve robot robustness.
• Write production-grade code (Python/C++) that integrates with our existing systems, using AI coding agents to accelerate progress and minimize entropy while avoiding AI slop.

We are looking for:

• Actively pursuing a PhD (preferred) or a research-heavy Master’s in Computer Science, Robotics, Machine Learning, or a related field.
• Expertise in training and deploying imitation learning models (VLAs/WAMs) for robotics with strong experimental hygiene and clear knowledge of the state of the art and where the community can improve.

Nice to have:

• Good grasp of machine learning best practices, architectures, and scaling laws among different domains.
• Experience troubleshooting and deploying algorithms on physical robot platforms, especially mobile and humanoid form factors.
• Experience working with and contributing to large-scale datasets (e.g., EgoScale, Open X-Embodiment) or specialized data collection approaches like Universal Manipulation Interface.
• Experience with large-scale cluster training (Kubernetes, SLURM, distributed GPU training) and maintaining high-quality codebases.

The hourly pay range for this position is between $30-$70/ hour. Base pay will depend on multiple individualized factors including, but not limited to internal equity, job related knowledge, skills, education and experience. This range represents a good faith estimate of compensation at the time of posting.