About the job
Flexion Robotics creates the intelligence layer for next-generation humanoid robots, drawing on deep expertise in robot reinforcement learning. Founded by scientists from organizations like Nvidia and ETH Zürich, the team has rapidly advanced from early development to deploying real humanoid capabilities. The company is backed by international venture capital and operates from Zürich.
Role overview
The Simulation Software Engineer will take full responsibility for the simulation stack that drives robot learning at Flexion Robotics. Simulation is central to the company’s approach for scaling intelligence, and this position shapes both the fidelity and efficiency of the simulation platform. The work directly influences how robots learn and perform in real environments.
Main responsibilities
- Simulation stack ownership: Design, develop, and maintain the core simulation infrastructure for robot learning. Oversee its architecture, reliability, and ongoing evolution to meet team needs.
- Simulation fidelity: Build and refine models for contacts, actuators, deformables, and sensors. Work to reduce the sim-to-real gap, improve physical realism, and validate simulations using real-world data.
- Training throughput optimization: Enhance the simulation stack for large-scale reinforcement learning. Implement efficient multi-GPU and multi-node pipelines, and resolve bottlenecks to increase training speed.
- Photorealistic rendering and synthetic data: Extend rendering pipelines to support vision-based learning. Enable domain randomization and generate synthetic datasets for robust perception systems.
- Software-in-the-loop testing: Maintain infrastructure for validating policies and systems within simulation before hardware deployment. Support rapid iteration and help minimize risks during transitions to real robots.
- Collaboration: Work closely with engineers and researchers in reinforcement learning, control, perception, and hardware to align simulation with real-world performance and deployment objectives.
