Founding ML Engineer

About Gradient Robotics

Gradient Robotics is building intelligent robots for data centers and factories. We believe the fastest practical path to capable general-purpose robotics is through industry deployment. We're here to change the world with robotics.

We own and build our full robotics stack — designing the hardware, building the software systems, training ML models, and manufacturing through a global supply chain — for world-class iteration speed, reliability, and scale.

We've built 3 generations of robots with a team of 7, backed by top-tier investors and working toward pilots with the world's largest data center builders. Our team created and built the best-selling open-source humanoid robots in the US at K-Scale Labs and worked on generations of foundation software for Tesla Optimus.

The Role

As a founding ML engineer at Gradient, you'll train and deploy the models that make our robots useful in the real world. You'll work across the full learning pipeline — from data collection and curation through VLA pre-training and fine-tuning to RL-based self-improvement on real hardware. Your models won't live in a paper. They'll fold boxes, move pallets, and operate in environments they've never seen before.

What You'll Do

• Train and fine-tune vision-language-action models for manipulation on Gradient's robots

• Work on distilling policies trained in sim environments to real environments

• Constantly assess data quality and setup pre-training + post-training split to fine-tune the backbone and action heads

• Debug training failures, identify failure modes, and close the loop between model performance and data collection strategy

• Own the full cycle: data → training → eval → deployment → autonomous improvement

Background

• Deep Learning: PyTorch or JAX, distributed training, mixed precision

• Robot Learning: Imitation learning, diffusion policies, flow matching, or VLA architectures (π₀, OpenVLA, ACT, RT-X, Octo, etc.)

• Reinforcement Learning: Sim-to-real transfer, distillation techniques

• Simulation: MuJoCo, Isaac Sim, or Genesis

• Python: Comfortable writing research-grade code that ships to production

Nice to Have

• Experience training or fine-tuning large VLMs (PaLI, Gemma, LLaVA)

• Multi-modal data pipelines at scale (video, proprioception, language, tactile)

• Publications at CoRL, RSS, ICRA, NeurIPS, ICML, or ICLR

• Experience deploying learned policies on physical robots

At Gradient, your models ship to real robots daily. You'll watch them get smarter, more general, and more reliable as you close the loop between learning and deployment. If that excites you, we should talk.