The New Era of AI-Based Bus Lane Enforcement and Driver Behavior Monitoring

By design, Territo says the technology is not being billed as a parking enforcement tool, rather a tool that’s designed to aid in the flow of transit and that the system was created with nuance to only enforce rules when violations impact that flow, not just if they’re seen at all.

“Our systems are only capturing events when those events impede the operation of the bus and the transit network, and I think that's important,” Territo said. “You know, I don't think anyone believes that there are never times when a vehicle may need to stop at a bus stop or stop in a bus lane to discharge a passenger for a brief moment. Quite honestly, as long as that doesn't impede the operation of the transit network, it's not always a bad thing. Cities are complicated, crowded spaces.”

Territo also explains the benefit of operating a system like that is it requires less compute and data transmission and allows for the processing of incidents to be done on the bus instead of having to be sent to the cloud for analysis. Once the on-bus system sees what it deems as a violation, it then creates a package of information and sends only that information for review instead of a constant stream of the camera feeds. Further, the AI isn’t just left to figure it all out on its own, as cities work with Hayden AI to chart out what constitutes a violation in pre-defined zones.

“Every program begins with a discussion of the rules of that particular city,” Territo said. “Our system is designed so that how a city treats a violation or a specific zone is specific to how our system detects events. So, if a city has certain hours that a bus lane is enforced, that is something that our system accounts for. If the system has certain rules with the size of a bus stop, that's something that our system accounts for. If our system has rules with the amount of time a vehicle can stop at a bus stop, that's something that our system accounts for.”

Territo explained that the first step was defining the rules, then applying those to a layer of mapping. The program then merges that mapping layer with a layer that detects vehicles, and when deployed, determines if passed vehicles fall within the parameters of a citable offense as defined by those rules. If the system determines it does meet the parameters, that’s when an event is created, and data is collected for the ticketing authority.

Since launching last April, 63% of registered vehicle owners who received one violation from the Philadelphia Parking Authority have not received a second ticket—a number that stretches to 91% when looking at the Metropolitan Transportation Authority’s multi-year implementation, according to Hayden AI. As recidivism numbers are low and dropping with subsequent check-ins—suggesting driver behavior is changing—the company doesn’t see this as a means to generate revenue from ticketing, but in cost savings that can be achieved through an increase in efficiency.

“So over time we would expect the benefits to shift and any benefit that comes from revenue, comes from revenue generated by more people riding buses, by less fuel being needed because of reduced idle time, by maybe less buses being needed because of faster, more reliable service and not because of just violation or fines,” Territo said.

As for reactions to the technology in practice, Zapson says it’s been a positive experience because of the equitable way enforcement is carried out—only ticketing when a violation impedes transit instead of just when a violation occurs at all.

“It's a happy surprise because there's always some caution when you're introducing a new enforcement tactic like this, but by and large, it's seemed pretty positive,” Zapson said.

In its implementation with Coach USA, Finch notes a 92% decrease in preventable incidents involving pedestrians, cyclists and motorcyclists since the company implemented Samsara’s dash cams in its coaches.

“Those numbers just are incredible, right?” Finch said. “I think when you imagine at that scale of over 1,000 buses being able to have that 92% decrease in these preventable incidents, it's just such an amazing proof point that we know so many other customer types can benefit from.”

As for the drivers, Finch says the platform is designed to keep them at the center, noting that feedback on the coaching was positive as it kept drivers actively aware instead of truncating training or review to a manager’s office. Finch also says this gives the opportunity to recognize and reward drivers for safe behaviors as the AI unlocks data points that managers may not have had time to calculate on their own in the past.

Finch notes that while she’s seen a lot of variety in the adoption curve with drivers, once a driver notices a change in themselves or has evidence from an incident they otherwise wouldn’t have, they become a “spokesperson internally” for their other drivers.

“Typically what we see is as soon as a tenured driver in the fleet has a moment where they either get exonerated from an accident or they notice a real change in their own behavior, like, ‘oh, I didn't realize I was picking my phone up that much,’ they can really help drive the buy-in and the adoption of the rest of the drivers,” Finch said.

As for how the cameras are making these detections, Finch explained that the AI is specifically seeking out things “on the edge” and using those to generate training events—only firing when it notices something it’s not supposed to see in normal driving. Further, the AI looks at the situation holistically instead of just focusing on one trigger to gauge the severity of the event.

For example, if the AI detects a mobile phone event, it then examines the rest of the available data from connected streams to assess things like road and weather conditions, vehicle information and GPS data to understand road characteristics such as speed and special zoning—like schools or construction—to assemble a full report. These events then turn out a personalized training item that aims to correct a behavior by delivering the why behind the lack of safety in the event determination instead of just reminding a driver to stay off their phone.

The approach of holistic data collection also allows for different events to be handled in different manners—small events can be recapped and noted in weekly AI training summaries, leaving more time and focus for larger events that need human intervention in training.

“We want to get people focused on the right thing and focused on impacting the biggest risk and the biggest impact drivers,” Finch said. “So, it's like, how can we help safety managers and coaches free up their time to focus where it really matters instead of just having a blanket approach for everyone, which won't work as effectively.”

The cameras are next set to be installed on 235 SamTrans fixed-route buses, expanding what the agency says is efforts to reduce collisions and improve safety for passengers, operators and the public.

“This technology gives our operators additional tools to stay aware of their surroundings while helping us better understand and prevent incidents,” said SamTrans Acting Deputy Director of Safety Omar Brown in a press release. “Improving safety for our riders, employees and everyone sharing the roadway remains a top priority for SamTrans.” 

What's emerging across both lanes of AI adoption is less a collection of individual products and more the early architecture of a connected fleet. A bus equipped with Hayden AI's ACE cameras can now coordinate with traffic signals through LYT. A bus running Samsara's suite of sensors and its AI-powered dash cam can share its data with maintenance systems, dispatch platforms and safety dashboards through an open API. In both applications, the technology is converging toward the same destination: a vehicle that doesn't just collect data, but acts on it—and increasingly, does so before a driver or a dispatcher has to ask.