Latest AI JobsBrowse Jobs
B

2027 Internship Behavior ML Engineer, Learned Manipulation Policies

bedrock-roboticsSan Francisco, CAEngineering
Apply on company site
Salary
Not listed
Job type
Internship
Experience
Intern
Posted
2 Oct 2026
Status
Open · checked 8d ago

About this role

Bedrock Robotics seeks a Behavior ML Engineer intern in San Francisco, California, to train learned manipulation policies for heavy construction equipment. This position suits students or researchers pursuing a degree in computer science, robotics, or a related field who want to apply machine learning to physical-world autonomy.

Role Overview

The Behavior ML team at Bedrock Robotics designs learned policies that dictate the actions of heavy machinery like excavators. As an intern focusing on learned manipulation policies, the candidate will train diffusion models and build evaluation frameworks alongside experienced engineers. The role involves addressing the multimodal motions of heavy equipment using real fleet demonstration data and simulation tools.

Key Responsibilities

Train and iterate on diffusion, flow matching, and related behavior policies utilizing real fleet demonstration data and simulated rollouts.
Conduct architecture, conditioning, and hyperparameter explorations to generate actionable findings for the team.
Develop and enhance evaluation pipelines that measure behavior models on task success, safety margins, smoothness, and operator-likeness.
Analyze failure modes such as distribution shift, mode collapse, and out-of-distribution scenes to suggest practical fixes.
Collaborate with simulation and evaluation teams to ensure offline metrics align with actual on-machine performance.
Write clean, reproducible, and reviewed code for the shared training codebase.
Report findings regularly to the autonomy, controls, and behavior teams.

Requirements

Currently enrolled in a BS, MS, or PhD program in machine learning, robotics, computer science, or a related field, or possessing equivalent industry or research experience.
Proficiency in Python and practical background in training models using PyTorch or equivalent frameworks.
Familiarity with generative modeling concepts like diffusion models, VAEs, or flow matching, alongside imitation learning or behavior cloning.
Practical experience executing and analyzing training experiments to distinguish real improvements from statistical noise.
Strong communication skills to clearly convey methodology, outcomes, and next steps.

Nice to Have

Published research or significant project background in imitation learning, robot learning, diffusion, or flow matching.
Background in training Vision Language Action models.
Practical history training reinforcement learning policies for robotic manipulation.
Background in evaluating policies in simulation and analyzing the sim-to-real gap.
Knowledge of experiment tracking tools and large-scale training infrastructure.
Previous exposure to physical-world control problems, autonomous vehicles, or robotics.
Interest in heavy equipment, construction, or earthwork.

Skills and Tools

Python
PyTorch
Diffusion models
Flow matching
VAEs
Imitation learning
Behavior cloning
Vision Language Action models
Reinforcement learning
Simulation tools

About the Company

Bedrock Robotics deploys autonomous systems on heavy construction equipment to improve job site safety and accelerate infrastructure projects. The organization has raised 350 million dollars and achieved initial fully autonomous excavator deployments in the construction sector within two years of operation. The staff includes industry veterans who contributed to projects like Waymo, Uber Freight, and Segment.

How to Apply

Interested candidates should apply directly on the official company careers page.

Skills and tech stack

PythonPyTorchDiffusion modelsFlow matchingVAEsVision Language Action modelsReinforcement LearningMachine LearningDiffusion

Questions candidates ask

What is the location of this internship?

The internship is located in San Francisco, CA.

What qualifications are required to apply?

Applicants must be currently pursuing a BS, MS, or PhD in computer science, robotics, machine learning, or a related field, or bring equivalent research or industry experience. Candidates also need strong Python skills, experience training models in PyTorch or equivalent, understanding of generative and imitation learning, and clear communication skills.

What kind of machine learning models will the intern work on?

The intern will work on diffusion flow matching and related behavior policies, generative models, imitation learning, and behavior cloning.

Is prior experience in construction required?

No prior experience in construction, earthwork, or heavy equipment is required, though an interest in the domain is preferred.

How should I apply for this position?

Applicants should submit their application on the official company careers page.

Ready to apply?

The application is on the official bedrock-robotics careers page.

Apply on company site