Hidden test drivers bamboozle pedestrians, as cars gauge reactions

An apparently driverless Ford van caught NBC’s attention this week, with the test-driver hiding from view by wearing a car-seat disguise. Thinking it had caught Ford in violation of current regulations, the news network investigated and unearthed the costume, which is apparently being used to study pedestrian behavior and reactions to cars without drivers.

Spotted driving in Arlington, Virginia, News4’s Adam Tuss tracked it down this week to take a look inside, discovering that its test driver was dressed, from the waist up, as a car seat. Able to fully control the Transit Connect, the driver’s garb thoroughly confused Tuss, who likened it to a sports mascot’s costume.

News4, an NBC affiliate, spoke to the Virginia Tech Transportation Institute, which confirmed that the Transit was part of a driverless car study. This isn’t the first time such a test has been carried out, but it has gained traction online because of the comedic effect of the reporter’s discovery – mild virality, if you will.

It comes in the same week that Intel has announced it will be deploying Level 4 SAE compliant vehicles by the end of the year, using its Mobileye wing – capable of completely autonomous driving, but still retaining space and physical controls for human drivers. The fleet will scale to 100 cars, apparently, and be deployed in the US, Israel (Mobileye HQ), and Europe.

In April, researchers from the University of California, San Diego (UCSD), used a disguise to monitor pedestrian reactions to a car without a driver – as part of a test to see how autonomous cars could communicate their intent to other road users and other passengers, according to professor Jim Hollan.

Speaking to the San Diego Union Tribue, Hollan added “many of the same issues about how people and robotic systems communicate arise in domains ranging from the surgical suite to controlling devices in your home. We are interested not only in communication, but also in trust. One wants an appropriate level of trust. Both too much and too little are problematic.”

It sounds like Hollan is anticipating the findings being used outside the automotive sector, in other automated machinery and systems – good news for other developers in the IoT, who might be able to use its findings inside their own applications.

The results of the experiment, which took place on the UCSD campus, found that many pedestrians did not notice the apparent absence of the driver – something mostly attributed to students “just walking without looking, while reading their cellphones,” according to Don Norman, head of UCSD’s Design Lab.

Norman believes that there would be different reactions off the campus, speaking to the Tribune, although people walking around glued to their smartphones seems a fairly universal habit. He notes that further research is hoping to better determine those different reactions, for use in self-driving systems.

“We want to know how people react to cars that have no drivers, so we’re testing cars that appear to have no driver. We need to know how everyone reacts, not just other motorists. We need to look at pedestrians, skateboarders, people on bicycles – people who use the roads,” Norman noted.

Self-driving vehicles will need to understand how pedestrians react, in order not to collide with them. Being able to tell when a pedestrian is about to step out into the road to avoid a pram or dog walker is a nice example, as is the ability to judge when a pedestrian is distracted (by a phone, child, jogger, dog, etc.), so that a vehicle can behave in the best manner – slowing to provide extra braking distance.

On the basis of the initial findings, the researchers are strongly recommending that an autonomous car should signal its intentions to other road users and pedestrians – ‘I’m about to move, I’m stopping, I’m about to turn.’ But the researchers are not mandating the UI or features that should be used to convey the car’s plans. envisions a roof-mounted panel that houses a screen, to spell out messages via images or text, which would let pedestrians know if the car was waiting to let them cross. Of course, cars already have a number of ways of conveying their driver’s intent, such as indicators (blinkers), flashing the headlights (which often means entirely different things in different countries), and the horn. Hand signals from the driver are also key.

Part of the future research will need to focus on how to integrate autonomous cars into the local ‘language’ of driving habits and gestures. For example, a UK driving dialect will differ from a French one, and developers will need to be able to accommodate for those differences – where flashing the headlights in the UK typically means ‘after you’ or ‘I’m letting you out,’ but effectively means ‘get out of my way’ in France. An additional example of potential complexity is that while the countries drive on opposite sides of the road, they typically yield to the right.

Returning to the aforementioned Intel project, Mobileye CEO Amnon Shashua noted that “geographic diversity is very important as different regions have very diverse driving styles, as well as different road conditions and signage. Our goal is to develop autonomous vehicle technology that can be deployed anywhere, which means we need to test and train the vehicles in varying locations.”

The automakers won’t like the idea of adding roof-mounted panels to their sleek designs, but will need to work on communication methods with pedestrians – often the most unpredictable road-users. Visual cues are the most likely candidate, but audio cues like chimes could be useful too – potentially allowing the cars to come across as friendly, instead of T1000-esque (silent killing machines, potentially poised to strike).

It is unlikely that an industry standard for this kind of unconnected and proximity communication is going to emerge any time soon, and of course, the whole issue could be avoided if pedestrians weren’t so skittish and unpredictable. A marketing campaign could go some way to alleviating the chance of an incident, but the mantle of ‘first driverless car that crushes a pram’ is one that the automakers will be avoiding like the plague. As such, these research efforts are vital for the industry.