Robotics Software Engineer Vision-Language-Action Models

Robotics Software Engineer Vision-Language-Action Models

Location

Trondheim, Norway or San Jose, California, US.

Role

As a Robotics Software Engineer specializing in robot learning and manipulation, you will drive the design, training, validation, and deployment of contact-rich manipulation skills on real robots. Your focus will be on solving dexterous, contact-rich manipulation tasks. You will develop and integrate sensor-based control strategies and Vision-Language-Action (VLA) policies, managing the full model lifecycle, from architecture and large-scale data collection through training, validation, and reliable deployment on industrial and collaborative robots.

This is a hands-on engineering role. You will spend significant time in the lab, running trials on hardware, curating high-quality datasets, and getting models to work on real robotic systems.

How you'll move the mission forward

  • Develop, train, and deploy multimodal VLA and learned manipulation policies that solve contact-rich tasks, integrating sensor-based control strategies that combine vision, force-torque, and tactile feedback.

  • Run the manipulation model lifecycle, conducting regular trials on real hardware to evaluate algorithmic changes and curate high-quality training datasets.

  • Architect and build modular components for imitation and reinforcement learning, continuously improving the robustness of contact-rich tasks.

  • Build pipelines for continual learning, enabling policies to keep improving from new demonstrations and real-world deployment data over time.

  • Stress-test and optimize the real-time execution framework for learned policies running on the robot.

Skills you will need to be successful

  • A Master’s or PhD degree in robotics, computer science, machine learning, or a related field, or equivalent professional experience in robot learning and manipulation (e.g. visuomotor policy learning, imitation learning, applied reinforcement learning,, Vision-Language-Action models).

  • Professional experience in Python and C++, with a proven track record of shipping production-quality code.

  • Deep expertise with a modern ML framework (PyTorch preferred; JAX or TensorFlow also welcome).

  • Direct experience testing and iterating on physical robots integrated with vision, force-torque, and tactile sensors.

  • Experience with robotic simulation environments (MuJoCo, Isaac Sim, Drake, or similar).

  • You thrive in the lab, spending significant time running hardware experiments and troubleshooting real-world edge cases.

  • Business fluency in English.

Skills that will differentiate your candidacy

  • Experience manipulating and analyzing complex, large-scale, high-dimensional multimodal data from varying sources, with proficiency training models at scale on cloud-based workflows.

  • Experience with edge deployment and model optimization for real-time control: low-latency inference, quantization, distillation, and pruning to meet tight control-loop rates on resource-constrained, on-robot hardware and accelerators.

  • Contributions to robot learning through publications or open-source projects.

  • Familiarity with ROS 2 and hands-on experience using industrial and collaborative robots from vendors such as Universal Robots, ABB, Fanuc, and KUKA.

  • Experience with multibody dynamics simulators such as Adams.