Physical AI Deployment Strategist
Technical Program Manager responsible for standing up and running a robotics training data collection program end-to-end—from rig fleet management and calibration to QA, dataset delivery, and reporting to model teams—using SQL/Python for dashboards and working hands-on with UMI grippers, GoPros, and VR headsets.
- Compensation
- $150K–$250K / yr
- Location
- Emeryville, CA
- Work setup
- Onsite
- Seniority
- Not specified
- Employment
- Contract
- Experience
- 3+ years
- Category
- Solutions engineering
- Posted
Skills
Must have
- 3–5 years as a PM or TPM with exposure to hardware, data pipelines, or physical workflows
- Proven ability to build data collection or technical operations programs from the ground up
- Comfortable owning task specs, QA programs, and standardization
- Hands-on experience calibrating rigs on the floor and writing stakeholder updates at a desk
- SQL or Python proficient enough to pull own data and build own dashboards
- Track record tracking and improving program metrics (usable hours, mix accuracy, SLA attainment, rejection rates)
- Experience providing technical input in a cross-functional environment with researchers and engineers
- Collaborative by nature, considering downstream impacts on operators, model teams, and infrastructure
- Bias toward action, scrappy problem solving, and moving fast without sacrificing rigor
Nice to have
- Experience supporting vendor or hardware supplier relationships
- Exposure to robotics, VLA models, or foundation model training data
- Experience scaling a data collection program from prototype to high-volume production
- Familiarity with computer vision, sensor calibration, or robotics data pipelines
Full description
About Us
Foundry Robotics is building an AI-native robotics manufacturing company focused on deploying advanced assembly and production capability for leading robotics companies and national-security-critical hardware. Basically, we’re building robots that build robots.
We are reimagining manufacturing through advanced robotics. Our mission is to rebuild the American manufacturing industry as an AI-first, assembly-focused, dual-use contract manufacturer. We aim to empower manufacturers with intelligent, efficient, and adaptable robotic systems that redefine productivity and quality.
The Role
This role is for a Technical Program Manager who thrives on turning model team requirements into production data, equally comfortable architecting QA systems from first principles and calibrating a rig on the floor. You will define what gets captured, how it lands, and whether it is usable, from early task spec through pre-delivery validation. You will stand up the data collection program from scratch and drive rapid build-test-learn cycles to deliver a dataset that is both rigorous and scalable.
You will be the day-to-day interface with our model team and the frontier labs we deliver to, and the source of truth for hours collected, hours delivered, and hours usable. You will partner closely with the Operations Manager, Robotics Data, who owns the workforce, floor execution, and daily fleet readiness.
Key Responsibilities
- Manage the rig fleet end-to-end (UMI grippers, GoPros, VR headsets and controllers), including procurement, lifecycle, asset tracking, spares, and vendor relationships
- Own rig calibration cadence and drift detection, including GoPro intrinsics, fisheye calibration, UMI gripper width drift, and per-operator VR eye-tracking calibration
- Define success criteria with the model team, translating high-level data needs into rejection rules an assembler can apply consistently
- Author SOPs and hand them to the Operations Manager for floor execution
- Own new task ramp-up, including spec, SOP, pilot batch design, and quality gate sign-off before full production
- Curate gold-standard demos (20–50 reference demos per task), maintained against drift
- Run post-capture QA (sampling + success classifier) and pre-delivery validation against model team spec
- Track diversity and coverage across task variants, object positions, lighting conditions, and operator styles
- Run daily capture logs and weekly dataset reporting to model and research stakeholders, including hours captured, hours usable, mix vs. target, top rejection reasons, and blockers
- Own task spec versioning, coordinating SOP updates, retraining triggers, and data v1/v2 marking when the model team changes a spec