About the role#
Teaching a 40-ton excavator to operate like an expert is a unique challenge. Unlike autonomous cars, these machines require multimodal motions where multiple approaches to digging or dumping are valid. The Behavior ML team develops the learned policies that guide these machines. We use diffusion policies to represent the full distribution of expert behavior rather than averaging it. As an intern, you will train these policies and help build the evaluation frameworks that determine their performance.
What you'll do#
- Train and iterate on diffusion-based behavior policies using real fleet demonstration data and simulated rollouts.
- Run architecture, conditioning, and hyperparameter explorations to produce clear findings for the team.
- Build and improve evaluation pipelines that score behavior models.
- Investigate failure modes such as distribution shift, mode collapse, and out-of-distribution scenes.
- Propose fixes for model failure modes.
- Work with simulation and evaluation teams to correlate offline metrics with on-machine performance.
- Contribute to the shared training codebase with clean, reviewed, and reproducible work.
- Share results regularly with the behavior, controls, and autonomy teams.
What you'll need#
- Currently pursuing a BS, MS, or PhD in computer science, robotics, machine learning, or a related field, or have equivalent research or industry experience.
- Strong Python skills and hands-on experience training models in PyTorch or equivalent frameworks.
- Working understanding of generative modeling, including diffusion models, flow matching, or VAEs.
- Working understanding of imitation learning or behavior cloning.
- Experience running and interpreting real training experiments.
- Clear communication skills.
- Published work or substantial project experience in diffusion policies, robot learning, or imitation learning is preferred.
- Experience evaluating policies in simulation and reasoning about the sim-to-real gap is preferred.
- Familiarity with large-scale training infrastructure and experiment tracking is preferred.
- Exposure to robotics, autonomous vehicles, or physical-world control problems is preferred.
- Interest in construction, earthwork, or heavy equipment is preferred.
Location & details#
- Location: San Francisco, California
- Term: Summer 2027
- Modality: On-site
- Employment: Full-time internship
- Compensation: Paid
About Bedrock Robotics
Bedrock Robotics develops autonomous technology for the construction industry. The company modifies heavy machinery to perform tasks with increased safety and precision. It focuses on infrastructure projects like housing, data centers, and energy facilities. Founded in 2024, the business operates out of San Francisco with 160 employees.
How to get in at Bedrock Robotics
Applying to Bedrock Robotics early gives you a distinct advantage because recruiters review applications before the candidate pool becomes unmanageable. Intern Insider sends an instant alert the moment a role matching your target is published, helping you apply among the first. This timing ensures your materials reach the team while they are still actively building their shortlist. You can reach out to recruiters directly to ask about the role or a referral, which often improves your response rate compared to a standard submission. Intern Insider surfaces the specific recruiters behind these roles to help you make that connection. Navigating the hiring process at a newer company like Bedrock Robotics requires this level of targeted outreach.
