Toronto’s first automated vehicle pilot: Robot safety on the roadway

Bern Grush, Executive Director

June 6, 2025

Is the Magna last-mile delivery robot delivering food more like a Starship or Serve robot delivering food or more like a Waymo or Baidu robotaxi delivering a human? Bern Grush, URF's Executive Director, explains what distinguishes PMRs operating on the road from those operating on sidewalks, and why the safety challenges are significant for every city considering this technology.

This spring, the Ontario Ministry of Transportation approved a Magna International Inc. application to operate a pilot for last-mile delivery robots in Toronto under the Province’s Automated Vehicle Pilot Program (AVPP). https://www.ontario.ca/page/automated-vehicle-pilot-program This is planned to begin in the second quarter of 2025.

Related to this, an invaluable Report for Action from Toronto's Infrastructure and Environment Committee is available at: https://www.toronto.ca/legdocs/mmis/2025/ie/bgrd/backgroundfile-254933.pdf.

Unlike the smaller, more commonly described robots that navigate city sidewalks and crosswalks, the Magna robots will operate on city streets with posted limits of 40 km/h or lower. Both of these types of mobile robots are referred to as public-area mobile robots [1] and are distinguished from most other mobile robots by the fact that they operate in public spaces among human bystanders that are uninvolved with the task of the robot, untrained in their operation, unprotected from their actions, and often inattentive as pedestrians and drivers are increasingly distracted.

Bystanders are context dependent

In the present case—because the Magna robots are constrained entirely to the roadway—bystanders are expected to comprise passing human drivers, nearby cyclists, and pedestrians using crosswalks. Pedestrians and vulnerable users on the sidewalk are far less dominant in this scenario.

Whatever critical duty of care Magna and its staff will take regarding bystanders—and I promise you it will be extreme because of its existential value to the business they hope to develop—these bystanders are not the greatest source of risk to automated vehicle road pilots. Jaywalkers are.

Recall that the first fatality of a vehicle under automated driving system (ADS) control was the 2018 death of a pedestrian jaywalker involving a crash during Uber ADS tests. That directly impacted Uber's Advanced Technology Group (ATG) and led to its eventual sale.

More recently, the jaywalking pedestrian dragging in San Francisco by a GM Cruise robotaxi, led to the collapse of that GM division, including the cancellation of its contract with Dubai to supply 4000 robotaxis by 2030 to that city’s public transportation system.

These two incidents triggered milestone losses to the fledgling robotaxi industry. They provide us with a key lesson: unstructured human behaviour represents the single greatest risk for the current state of automated driving systems on our roadways.

Road risks and sidewalk risks are very different

There are three critical PMR safety differences between the sidewalk and the road: speed, size, and rules. On an urban sidewalk, a mobile robot is moving among and spatially competing (some say collaborating) with pedestrians that are generally moving between 2 and 8 km/h while themselves usually constrained to 5 or 6 km/h. Pedestrians may be annoyed, robots may have to pull over and wait, but no one is getting injured.

On the roadway, a mobile robot moves within an assigned lane (unlike on the sidewalk), moves far more quickly than on the sidewalk (up to 32 km/h in the Magna-Toronto pilot case), and must follow provincial highway traffic rules which provide far more regulation of flow than is available for the sidewalk.

So, there is a trade-off: the roadway is far more structured, therefore easier to understand, but far less forgiving because of its greater stores of momentum. That is the critical reason we require you to be licensed to drive but not to use the sidewalk. Mistakes on the sidewalk are more forgiving.

Herein lies an important conundrum. Automated driving systems that operate on roadways must be able to flow among a large variety and volume of other motor vehicles that share these roadways.

There is no choice but to have the relevant senior government responsible for roadways—whether National, State or Provincial depending on your country— adjust its motor code as well as its safety, certification, licensing, and enforcement regimes to accommodate these systems.

But when we apply automated driving systems to devices that operate on sidewalks and bikelanes, the appropriateness and readiness for senior governments to adjust motor codes and take accountability for the rules of engagement in these spaces is often beyond the scope of senior transportation departments. Worse, when similar devices with automated driving systems operate within buildings such as airports, hospitals or shopping malls, the connection is yet harder to see.

The robot in the roadway

Let's return to the specific case at hand—700-pound robots travelling at a maximum of 32 km/h on streets limited to 40 km/h in a large city. Setting aside the unrelated mobility complexities of the sidewalk, consider five safety circumstances specific to this pilot.

Following a larger vehicle

While delivery robots navigating the sidewalk experience the greatest single safety threat when entering the crosswalk (due to right turns on red by motor vehicles), robots operating on the roadway would only encounter crosswalks as they pass through an intersection, or as they themselves turn right on red. Magna’s robots pass over crosswalks rather than traverse them.

Nonetheless, a specific intersection risk critically associated with a Magna-type PMRs is when following a larger vehicle through an intersection. When a considerably smaller vehicle follows a much larger vehicle, the smaller vehicle may easily be occluded from the sight lines of the driver in an opposing motor vehicle that is turning left. The left turning vehicle driver, if in a hurry or careless, risks clipping the robot if it is occluded until the last moment.

Jaywalkers

Jaywalking behaviour is common, sometimes careless, and often a source of “edge-cases” relative to robotic machine vision. As mentioned earlier, two substantial robotaxi companies, Uber ATG and GM Cruise, have already been taken out of the robotaxi industry by crashes with careless jaywalkers—each company enduring staggering losses.

While the safety risks are much lower for Magna's robots (due to their lower momentum compared to a robotaxi that is seven times the weight of a Magna robot), because the potential value of the subject pilot to each of Ontario, Toronto, and Magna is very high, this risk is existential at this early commercialization phase.

While the probability of an outlier jaywalker mishap should be very low during such a cautiously planned pilot, recall the recent “low probability” AV mishap in Whitby (p. 5 of the IEC Report for Action). While this incident did not involve a human jaywalker, my point is that outlier events will happen on the roadway, and jaywalkers are an important source of such outliers.

Three wheels

In addition to interactions with other road users, I also have mechanical road-safety concerns.

The 700 lb Magna robot runs on three wheels. The two front wheels on an axle are the powered wheels; the single back wheel is a support wheel.

A 1982 SAE technical paper, “Three Wheeled Vehicle Dynamics” (820139) indicates that among 3- and 4-wheel designs tested for that paper “the 3-wheeled vehicle with two wheels on the front axle” is the least stable. The full context is:

“Comparisons are made between a 3-wheeled vehicle with two wheels on the front axle, a 3-wheeled vehicle with two wheels on the rear axle, and a standard 4-wheeled vehicle. Each vehicle’s lateral stability, rollover stability during lateral acceleration, rollover stability while braking in a turn, and rollover stability while accelerating in a turn are determined. It is shown that for lateral stability, the 3-wheeled vehicle with two wheels on the rear axle is more stable than the 4-wheeled vehicle, which is in turn more stable than the 3-wheeled vehicle with two wheels on the front axle.” https://www.sae.org/publications/technical-papers/content/820139/

Accordingly, the Magna design may be relatively the least laterally stable among potential 3- and 4-wheel designs. Of course, I do not know whether Magna’s designers have defeated this problem. While I assume Ontario may have already examined Magna’s stability testing data, it would behoove the Province to ask for more during this pilot. Never let a good data collection opportunity go to waste.

I also empathize with the City, here. While this is a Provincial pilot, the City will necessarily be left to address crash-related traffic matters.

Single wheel

If the two front (powered) wheels were to straddle a pothole of a few inches in depth in such a way that the back wheel were to be caught in that pothole, this could make the robot unstable (sudden swerve, tip-over, or worse) at its middle or higher speeds, or if the wheel were torn off. This may not happen very often, and perhaps not at all during the pilot, but someday, it will.

Here is a short 13 sec video of the robot in action on dry pavement and without competing traffic: https://www.cbc.ca/player/play/video/9.6746599

Chase vehicles

A primary navigation focus for mobile robotics is "don't hit anything." Toward that concern, and in the case of this pilot, the plan is for humans in a chase vehicle able to stop the robot and have a human to push it to the side of the roadway. There is some comfort in that precaution from a test perspective, although using a chase car has its own problems such as adding more traffic, additional sight line occlusion to the smaller vehicle, separation at light changes, and risks to the human attendant that exits the chase car to enter the street and work in a potentially risky traffic space. We all know this arrangement is not sustainable, rather, my larger point is that there are also many things which the chase vehicle cannot prevent, but also need to be anticipated from the City’s and Province’s perspectives.

Why this matters

While it is reasonable to suggest that these robots are able to travel hundreds of kilometres without incident, these few issues outlined above, and many others outlined in the draft ISO 4448 standard series, would arise if such robots were to be deployed at scale for its intended applications on Toronto streets.

Clearly, Toronto needs PMR pilots and trials, and its City transportation staff may already be aware of many of these issues, but our full understanding is far from complete. No one’s is.

And that is why this pilot and many others like it are critical. Not only to my city and province, Toronto, Ontario, but to all the cities that will adopt and govern this technology 2025-2045.

As your city considers similar pilots, be sure to ask your representatives to explore all facets of this technology: safety, traffic patterns, insurance rates, job changes, accessibility opportunities, and many others.

Our free URF Executive Guide to PMRs is a great starting place. Download it here: https://www.urbanroboticsfoundation.org/executive-guide

[1] see ISO/TR 4448-1:2024 https://www.iso.org/standard/81068.html