Artificial intelligence advises INL buses

What would you say to receiving accurate weather and road conditions delivered to your vehicle in real time with milepost resolution? It’s not a pipe dream. It’s currently a reality for the Idaho National Laboratory (INL) bus system in a collaboration with IBM Watson’s Internet of Things (IoT) artificial intelligence (AI) system that’s been four years in the making.

INL buses drive 3,000 passengers every day from communities along the Snake River to the lab’s 890-square mile site in the Arco Desert, which locals just call “The Site.” The roads to the Site are among the least hospitable in Idaho, a smorgasbord of the worst weather conditions Idaho has to offer.

Because the roads can be so dangerous, the INL evaluates its bus routes on a daily basis. Road scouts travel the bus routes, especially during adverse weather conditions, to find hazardous conditions before any INL buses begin to pick up riders. If conditions are bad, buses may be rerouted or, in the worst case, bus service can be canceled.

Tad Pearson, department manager of transportation services for the INL, described to the Idaho Business Review what road scouting is like: “We have road scouting vehicles. So from watching the weather forecast…if we get a forecast that says we have a possibility for a bunch of snow, a bunch of wind, black ice, that kind of stuff, (then) we make a decision the day before that we’re going to scout the roads early the next morning.

“We take the main arteries: I-15, US-20, US-26, Highway 33. We drive those routes, before the buses are even up and running. And we determine at that point whether we can run those routes or not.”

Road scouting may sound easy but the distances are long just going one way on the main routes, around 160 miles total, 190 if the segment from Pocatello to Blackfoot is added in. The management team also considers the data from the Idaho Transportation Department (ITD) road weather information station (RWIS) network. Unfortunately, the RWIS locations are spread out, and in the Arco Desert, entire micro-climates can exist between camera locations.

Scott Wold, the director of mission support services for the INL, described the limitations of scouting and RWIS data: “At three o’clock in the morning you take a look at…(the) camera at Telegraph Hill (between Idaho Falls and the Site main gate), everything looks clear. You take a look at that camera at the INL Puzzle (the intersection of the Site main gate, US-20 and US-26 ), it looks clear. But in the middle, you’ve got an ice storm that you can’t see and that’s because there’s one of those little microclimates basically between (those locations).”

Before 2017, the road scouts would leave around 2:30 or 3 a.m. to drive the bus routes in bad weather with just RWIS data and a weather forecast. The IBM collaboration has taken most of the guess work out of evaluating the roads and will eventually make road scouting obsolete.

Wold attended a presentation in 2016 about IBM Watson’s AI that adjusted environmental building controls in real time, the aim of optimizing power usage and saving on utility costs. “I was sitting in on that presentation…and I thought, if you could do all that work for a building, then why not for a road? So I reached out to IBM and started partnering with them in 2017.”

The INL contracted with IBM to develop a weather and road condition prediction system using IBM’s AI. To help build the database to teach the AI, ITD worked with the INL to install sensors on three INL road scout vehicles. The data ITD collected on pavement conditions was uploaded in real time to a Vaisala Navigator website for use by INL dispatch and management. ITD also used the data to supplement its fixed RWIS sites.

The current system on the INL buses uses millions of data values to identify correlations between the mobile road scout/ITD data, historical RWIS data, National Weather Service observations and other data sources to predict road conditions in real time and to send that to every bus.

Real-time weather and road data is sent to IBM Watson’s supercomputer, Wold explained. “It runs 178 different weather models every 15 minutes and then takes the best of the best,” he said. “So we get an updated forecast every 15 minutes of what the road conditions are going to be like at every mile marker on I-15, US-20, US-26 and Highway-33; and it will do that really accurately up to 24 hours in advance.

Wold added: “What it’s doing is it will forecast, it will check (the predictions for accuracy), and then it will adjust. So every time you have a little storm come through, it gets smarter. The only way I can describe it is that it’s scary how accurate it is.”

The system in the INL buses is now robust enough that the INL and IBM are expanding it. In 2020, I-86 and I-84 between Idaho Falls and Boise were added. I-90, US-95 and US-20 are on the list of roads to add. ITD has remained close to the project and would like to add the AI’s capabilities to Idaho’s 511 system.

The INL transportation staff can look forward to retiring its scout cars sometime soon.

“Here’s another system that the national lab developed that can now be used anywhere,” Pearson remarked. “It’s kind of cool to be a part of that. From my perspective, what a great feeling that is, as somebody who has to get up at 2 or 3 o’clock in the morning, and be the only person on the road (to scout bus routes) and not know for sure what’s ahead of me or behind me…We have a system that could do that for us now and I won’t have to put my own safety at risk to be out there.”