Machine Learning Researcher
Droyd · San Francisco
About the Team
Droyd builds autonomous robotic systems that take on repetitive manual work in real environments. Our robotic arms run on tight compute and power budgets, so learning systems have to be fast, reliable, and deeply integrated with hardware.
Our AI team designs and ships the models that let robots see, reason, and act. This work runs on real machines, not benchmarks.
About the Role
As an AI Researcher at Droyd, you’ll own meaningful parts of the learning and inference stack that power our robotic arms. You’ll train models, push them onto hardware, and iterate until they work in the real world.
You’ll work in person with a small, senior team across robotics, controls, and data. Your work will ship directly to deployed systems.
This role is based in San Francisco, CA. We’re an in-person company. We build faster that way.
In This Role, You’ll
Architect and build training and inference stacks for models running on low-payload robotic systems
Design and train new model variants, run experiments, and document results
Propose and explore new research directions that improve speed, reliability, or capability
Develop fine-tuning and optimization methods tailored to robotics workloads
Improve throughput across the full training and deployment pipeline
Work with the data team to manage datasets and keep pipelines clean
Deploy models to hardware and debug real-world failures
We’re Looking For Candidates Who
Has experience with modern ML frameworks like PyTorch or JAX
Understands vision-language models and how they behave in practice
Holds a Master’s, PhD, or equivalent hands-on research experience in ML, AI, or CS
Can ship clean research code and reason clearly about model behavior
Has interest in robotics, controls, or embodied AI
Bonus: experience with edge-device inference or real-time constraints
About Droyd
Droyd builds autonomous robotic systems to automate manual work for enterprises. We design the hardware, write the control stack, collect our own data, and train models that run under real-world constraints.
If we do this right, robots stop being demos and start being tools people rely on every day.
Join us and help build systems that actually ship.