About Genki Robotics

Genki Robotics was founded by a team of pioneers in humanoid robotics committed to practical applications of physical AI. With decades of collective expertise, we are developing advanced robotic systems that combine intelligence, dexterity, and resilience to address critical challenges in public safety, urban maintenance, and other mission-driven environments. Our vision is to accelerate the integration of intelligent humanoids into society, augmenting human capability and shaping the future of work and service.

We’re backed by a16z, AMD, DCM, Incubate Fund, and X&KSK.

About the Role

You will own the software that turns demonstrations into training data. Our models are only as good as what we feed them, and the path from a human operating a robot to a clean, aligned, training-ready episode has more failure modes than anyone expects.

This is an infrastructure role at the center of how we learn new behaviors. When data is the bottleneck, you are the one who unblocks it.

What You'll Do

• Own the capture pipeline: time-synchronized recording of video, depth, proprioception, force, and action streams into coherent episodes.

• Own retargeting from operator input to robot kinematics, so demonstrations are reproducible by the policies trained on them.

• Get action timing right, including the delay between operator intent and the command taking effect on the robot.

• Build validation and automated rejection so bad episodes are caught at capture time rather than discovered months later in a training run.

• Build the tooling that makes collection productive: replay, annotation, labeling, and dashboards that show whether a session was any good.

• Own throughput and quality metrics, and drive usable episodes per operator hour.

• Serve data to training: dataset assembly, versioning, and the interfaces the ML team works against.

• Work with Behavior Learning to shape what gets collected, and close the loop when a policy fails because of data rather than modeling.

What We're Looking For

• Strong systems engineering in real-time or robotics software. Python and C++.

• Experience with multimodal sensor data: synchronization, timestamping, buffering, and the ways these break silently.

• Experience with robotics middleware and the mechanics of getting data off a robot reliably.

• Track record building tooling other engineers depend on daily.

• Judgment about data quality, and the instinct to check whether the data is wrong before assuming the model is.

• Comfort working close to hardware, in the lab, with a real robot.

Bonus Points

• Teleoperation systems, including latency, feedback, and operator ergonomics.

• Kinematic retargeting from human motion to robot embodiment.

• Experience building datasets for imitation learning.

• Data infrastructure at scale: storage, versioning, and lineage for large multimodal datasets.

• Autonomous vehicle data collection or fleet logging experience.

Compensation: Competitive base + equity