Member of Technical Staff, Generative Policies

Baseline Robotics; Zürich; Vor Ort; CHF 120.000 - 170.000; Vollzeit

Benefits dieser Stelle

Stock options

Zusammenfassung

Baseline Robotics in Zurich, Switzerland, is seeking an experienced engineer to advance diffusion-based policies on real hardware. You will focus on generative modelling for sequential decisions and publish impactful results.

You will work with Python, PyTorch, and GPU-accelerated simulators such as Isaac Lab, Newton, mjlab, Genesis, with opportunities to publish in top venues. This role is on-site in Zurich with stock options included.

Qualifikationen

• Experience training and evaluating generative policies on real hardware.
• Knowledge of generative modelling for sequential decision problems.
• Publications in top machine learning or robotics venues, or equivalent experience.
• Fluency in Python and PyTorch.
• Experience with JAX or Rust.
• Experience with GPU-accelerated simulation tools (Isaac Lab, Newton, mjlab, Genesis).

Aufgaben

• Train diffusion and flow-matching policies on demonstration data.
• Tune action representation and chunking on real hardware.
• Condition policies on visual history, proprioception, and language.
• Choose sampling to preserve action multimodality within the control budget.
• Recover from the off-distribution states a policy reaches by acting on prior outputs.

Kenntnisse

Python & PyTorch

Generative modelling

Training on real hardware

Publications

JAX or Rust

GPU-accelerated simul tools

Tools

Isaac Lab

Newton

mjlab

Genesis

Jobbeschreibung

About Baseline

Open source revolutionised how we write software, but its impact on robotics stays limited due to expensive hardware accessible only to well-funded labs. We run that hardware as shared infrastructure for the broader community, billed only for active compute. For a given task, policies are evaluated under similar conditions, giving researchers common ground for fair comparisons.

In this role you will

• Train diffusion and flow-matching policies on demonstration data.
• Tune action representation and chunking on real hardware.
• Condition policies on visual history, proprioception, and language.
• Choose sampling to preserve action multimodality within the control budget.
• Recover from the off-distribution states a policy reaches by acting on prior outputs.

What we hope you'll bring

• Experience training and evaluating generative policies on real hardware.
• Knowledge of generative modelling for sequential decision problems.
• Publications in top machine learning or robotics venues, or equivalent experience.
• Fluency in Python and PyTorch.
• Experience with JAX or Rust.
• Experience with GPU-accelerated simulation tools (Isaac Lab, Newton, mjlab, Genesis).

We encourage you to apply even if you do not believe you meet every qualification.

Details

• Location: Zurich, Switzerland, on-site 5 days a week.
• Annual salary: CHF 120,000 - 170,000.
• Every full-time hire receives stock options as part of the stated salary.

Deadline to apply: None. Applications will be reviewed on a rolling basis.