Oversight for a routing AI
Led research and interface design for ARTC’s Hunter Valley network control, moving train controllers from manual routing to supervising GE’s routing AI.
Mudbath · ARTC / GE · 2017
Every train in the Hunter region was planned by hand, on paper. It took an enormous amount of time and skill from train operators, on top of managing maintenance requests and emergencies.
ARTC’s Hunter Valley network control program (ANCO) brought in GE’s routing AI to plan train movements. The plan could come from the machine, but controllers still carried responsibility for the network. They needed to see what the AI was doing, and to step in when something didn’t fit.
I led the operator research and interface design. That meant understanding how controllers actually work through demanding four-hour shifts, and designing for that reality rather than for a demo.
The interface gave controllers oversight of the AI’s routing, and a way to steer the exceptions it couldn’t judge on its own, such as priority trains and dock loading. It ran on the web, so operators could plan from any device, at any time.
We validated it with usability testing at every stage: wireframes, prototypes and visual design.
Controllers moved from routing trains by hand to supervising the AI, and the work was delivered with positive operator feedback.
“It was the question I still work on: when should a person hand over to the machine, and how do they take back control when they shouldn’t?”