Machine Learning Researcher at Droyd in San Francisco, CA

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 someone 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.