Role Overview
You will develop perception, manipulation and autonomy software for humanoid robots and close the loop between models and real-world execution. This includes fine-tuning models, building imitation & simulation training pipelines, productionising model and software deployments, and operating robots in customer contexts (on-site testing, teleoperation and handovers). You will be hands-on with robots daily, iterate fast with hardware in the loop and deliver robust, safe systems for customers.
Key Responsibilities
- Develop and integrate perception → action pipelines (vision models, VLA, grasping and manipulation) and productionise them onto robot platforms
- Experience in training and fine-tuning deep learning models
- Design, run and iterate imitation-learning and simulation training workflows to produce robust policies and manage sim→real transfer
- Work on on-robot systems (Ubuntu / NVIDIA Jetson), battery/teleoperation flows and debugging sensors, actuators and hardware integration
- Maintain CI/CD and reproducible deployment pipelines for models and robot software (Docker, GitHub Actions / GitLab CI or equivalent)
- Produce clear technical documentation and runbooks (documentation-as-code mindset) and collaborate with customer teams during on-site installations and handovers
Must-have (Required)
- Fluent in English with excellent verbal and written communication
- Comfortable working in Scrum sprints and delivering in short cycles
- Strong hands-on experience with ROS / ROS2 development and integration
- Practical experience with Linux / on-robot systems (Ubuntu) and edge compute platforms (NVIDIA Jetson or similar)
- Practical experience building or integrating perception & manipulation systems (object detection, grasping, closed-loop control)
Strong Pluses (Nice-to-have)
- Experience with MLOps, dataset pipelines and local training infrastructure for robotics / vision models
- Experience with simulation + imitation learning workflows and sim→real transfer
- Cloud & infra skills (AWS / GCP, Kubernetes) and experience operating fleet-style services (message queues, event streams)
- Prior experience with customer projects, field installs or customer onboarding
What Success Looks Like
- Models reliably detect and localise objects and the robot successfully grasps and manipulates target items in realistic customer environments
- CI/CD pipelines and MLOps workflows enable repeatable model training and safe rollouts to robots with monitoring and rollback
Culture & Soft Skills
- Customer-facing pragmatism: you can translate prototypes into maintainable, testable customer deliverables and support field installs
- Documentation first: you write runbooks and docs alongside code and practice documentation-as-code
- Collaborative, curious and pragmatic — you iterate fast with hardware in the loop and help grow a small, SOTA team
