Revolute Robotics

A robot that drives and flies

Inside a pressure vessel, a cooling tower or a run of pipe, there is no GPS, no room to manoeuvre, and no good reason to send a person. The robot drives when the floor allows it and flies when it does not, and it is built to survive the walls it is inevitably going to hit.

The platform went through three bodies while I was there. HMR is the caged drone, wrapped in a rolling cage so a collision is a bounce rather than a crash. TQ dropped the cage. GRIFFIN moved to ducted rotors and is the current version, and it is the one whose autonomy I owned.

I joined as a control systems engineer and left as the senior robotics engineer, and the arc of the job was the whole path from a bench prototype to hardware that customers ran on their own sites.

Flying inside the plant

This is the part that is hard to argue with. The robot holding position and moving through a live industrial structure, walls on every side and no satellite fix anywhere in the building.

Knowing where you are with no GPS

Everything above depends on the robot knowing its own position, and it cannot ask a satellite. I led integration and algorithm development for a LiDAR-IMU localization and mapping stack that holds cumulative pose drift to 1 cm over a 100 m traverse, with loop closure landing in 50 ms.

One centimetre over a hundred metres is the difference between a map an inspector can act on and a map that is merely a nice picture. The perception stack fuses LiDAR SLAM, visual SLAM, micro-ToF and RGB; planning and control run on ArduPilot and MavROS.

Field crews needed to trust it too, so I put real-time point cloud visualisation and map quality feedback into the operator workflow. Inspection coverage gets verified before anyone packs up and drives home, not a week later at a desk.

Rebuilding the platform

Before any of that worked, it had to be made to work. In my first month I rebuilt eight subsystems across software, firmware, control, sensing and the mechatronic architecture, then took the platform through five hardware and software revisions.

Two changes mattered most. I wrote custom firmware for an MPC-based locomotion mode and tuned the PID gains against logged data, cutting energy per manoeuvre by 30%. Then I re-engineered the powertrain across motor drives, gear ratios and battery management for a 30% better thrust-to-weight ratio, which is what buys payload and endurance on a robot that has to lift itself.

Out of the lab

A robot that works in the lab is a demo. I led field deployments to customer sites, including a coal mine outside Phoenix. Pre-deployment validation, flight testing on site, and diagnosing failures on someone else’s schedule with the crew standing there watching.

Across 30 days of field operation the robot flew reliably inside live tanks, towers and pipe runs.

Shipping a fleet

Production is its own engineering problem. I owned project planning and the production ramp for the fleet, coordinating a five-person build team across seven units and cutting build-to-field turnaround by three weeks.