Are you passionate about using reinforcement learning, behavioral cloning, physics simulation, and humanoid robots? Curious what you’d be able to accomplish with total access to Boston Dynamics robots? As a Research Engineer on the Atlas Behavior Learning team, you will join a world-class team of engineers and scientists focused on creating groundbreaking mobile manipulation behaviors for humanoids.
We are investing in reinforcement learning, behavioral cloning and physics simulation as a key technology for achieving dexterous and robust whole-body manipulation that can be deployed in real-world environments.
In this role, you will be responsible for:
Train RL and BC policies for challenging bi-manual manipulation tasks
Write well documented control code that enables deployment of trained policies to Atlas
Lead development of evaluation frameworks that enable quantitative benchmarking of policies
Work on scalable simulation and data management tools for storing unstructured training data
Collaborate with research scientist on state of the art reinforcement learning and behavioral cloning algorithms
We are looking for:
MS in Computer Science, Machine Learning, Robotics, or a related field
Excitement for integrating, running, and evaluating their learned policies on the robot
Experience training, deploying and debugging RL policies on real robots
Strong foundation in Python, C++ and modern numerical frameworks (e.g., PyTorch and Jax)
Experience in databases, compute clusters, kubernetes, docker and CI/CD
The ideal candidate has:
Extensive experience with training RL policies in simulation with domain randomization
Experience with working in large multi-language code-bases
Experience with writing real-time control software for robots
Why join us?
Direct access to cutting-edge robots and the infrastructure to run large-scale experiments
A collaborative, mission-driven team where your ideas have real impact
The chance to help define what’s possible in real-world robotics
#LI-JM1
Top Skills
What We Do
Boston Dynamics builds advanced mobile manipulation robots with remarkable mobility, dexterity perception and agility. We use sensor-based controls and computation to unlock the potential of complex mechanisms. Our world-class development teams develop prototypes for wild new concepts, do build-test-build engineering and field testing and transform successful designs into robot products. Our goal is to change your idea of what robots can do.
