Robotics Engineer

Robotics Engineer, Imitation Learning and

Autonomous Manipulation

FacilityOps.AI is hiring a robotics engineer in Kenya to teach robot arms precise,

repeatable tasks from human demonstrations, then make them run autonomously and

safely. Your first target is automated fiber-optic connector inspection and cleaning forAI

data centers.

About us and the work

FacilityOps.AI builds robots that inspect data centers and critical facilities: electrical

rooms, cooling and network rooms. Our robots walk fixed routes with thermal, visual and

LiDAR sensors and produce a verified record of what they find.

We're now adding manipulation. AI data centers need tens of thousands of fiber

connections inspected and cleaned before they go live, and there aren't enough trained

technicians to do it. You'll help build:

A bench station where an arm inspects fiber end-faces under a microscope, cleans

them and logs pass/fail.

An in-rack robot: an arm on a mobile base that removes dust caps and cleans ports in

racks before turn-up.

Door and panel skills for our inspection robots: opening a cabinet, inspecting inside,

and closing it again.

Location Kenya (Nairobi preferred), Hybrid

Type Full-time

Pay USD 1,750 per month, based on experience

Reports to Robotics lead (US)

Team: 1 US robotics lead, 2 engineers in Kenya, 2 in

You'll work mainly with a Kinova Gen3 arm, Hello Robot Stretch mobile manipulators, ROS

2 and PyTorch.

What you'll do

Imitation learning: train manipulation policies from human demonstrations (for

example ACT, Diffusion Policy, or fine-tuned vision-language-action models) for tasks

like gripping a connector, removing a dust cap, inserting into a cleaner, or opening a

latch.

Demonstration data: design how demos are collected by teleoperation (leader-

follower arms, VR or SpaceMouse), write collection guidelines for the US team, and

clean, label, and version the datasets.

Autonomous manipulation: combine learned policies with classical control: motion

planning (MoveIt 2), grasp and pose estimation from depth cameras, and force/torque-

aware insertion.

Simulation: build and maintain simulated versions of the tasks (MuJoCo or Isaac Sim)

for testing, data augmentation, and sim-to-real transfer.

Safety and reliability: force limits, no-contact zones, failure detection and automatic

stop-and-recover. Measure success rates over hundreds of trials and report them

honestly.

Deployment: package policies as ROS 2 nodes that run on the robot's onboard

compute, and debug them remotely with the US lead on real hardware.

Local testing: run early experiments on a low-cost arm we ship to you in Kenya (for

example, a LeRobot SO-101-class arm) before moving to the Kinova.

What you need

Must have

Degree in robotics, mechatronics, mechanical, electrical, or computer engineering, or

computer science, or equivalent hands-on experience.

2+ years building robot manipulation, including at least one project you took from data

to a working policy on a real arm (research, competition,or industry).

Hands-on imitation learning: trained and evaluated at least one behavior-cloning policy

(ACT, Diffusion Policy, or similar) with PyTorch.

JD: Robotics Engineer, Imitation Learning (Kenya)

ROS 2 (nodes, topics, actions, TF, launch files) and MoveIt 2 or a similar motion-

planning stack.

Strong Python; working C++.

Robot kinematics, coordinate fram,es and camera calibration.

RGB-D perception for grasping: object pose estimation, point clouds, OpenCV.

Linux, Git, Docker, and training on cloud GPUs.

Clear written English. You'll work remotely with US and Pakistan teammates.

Nice to have

LeRobot, robomimic, OpenVLA, pi0 or other vision-language-action models.

Force/torque control or contact-rich insertion tasks (peg-in-hole, connector mating).

MuJoCo or Isaac Sim, and sim-to-real transfer.

Kinova, Franka, UR or Hello Robot Stretch hardware.

Building teleoperation rigs (leader-follower arms, VR).

Fiber optics, data centers or electronics assembly.

Published papers, open-source contributions or demo videos of your robots.

1. Online exam

2. Technical interview: walk us through a manipulation or imitation-learning project you

built.

3. Take-home task (paid or unpaid, about 6 hours): train and evaluate a behavior-cloning policy on a

provided simulated or open dataset, and write up what worked and what didn't.

4. Final call with the team.

<