Self-Driving Vehicles: Will we have to go through a semi-autonomous stage?

By June 07, 2016
futur de la mobilité

Interview with Régis Vincent, Head of Software at SRI Robotics. He talks about Yamaha’s flagship Motorbot project and gives us his predictions as to when self-driving cars and household robots are likely to become commonplace.

We met up with Régis Vincent, Head of Software at SRI Robotics, a unit of SRI International, which is a non-profit, independent research centre serving government and industry. Located in Menlo Park at the heart of Silicon Valley, SRI International runs projects for government agencies, notably the Defense Advanced Research Projects Agency (DARPA) – the US Ministry of Defence agency which develops technologies for military use – as well as private sector players, both large firms and startups – intended to develop disruptive innovations. The stated aim of SRI International is to move R&D from the laboratory to the marketplace. 

SRI is hardly a household name, but the organisation has nevertheless been behind a large number of the devices which we now use in our daily lives. Since the research centre was founded 65 years ago, its engineers have been closely involved in the development of such innovations as colour television then colour photographic film in the 1950s, ultrasound for medical diagnostics in the 1980s, computers as we know them today, Arpanet, a 1960s precursor to the Internet, and more recently Siri – the first-ever virtual personal assistant, later acquired by Apple.

Régis Vincent devant les locaux de SRI Robotics

A longstanding contributor to L’Atelier Numérique (L’Atelier Digital), Régis Vincent was talking to us about robotics as long ago as 2004 and in 2010 he introduced the idea of the ‘robot car’, the elder brother of the self-driving car which we hear so much about today. This robotics expert, who gets up close and personal on a daily basis with the technologies that will be available to us all in ten years’ time, tells us about the Motorbot project, which was unveiled at the Consumer Electronics Show in Las Vegas earlier this year, and gives us his thoughts on when self-driving cars and household robots are likely to become commonplace.

Could you tell us about the Motobot project? What potential applications does this technology have?

Régis Vincent : Well, Yamaha approached us a little over a year ago to work on the Motobot project. The idea was to develop a humanoid robot capable of riding a motorcycle at high speed without having to modify the structure of the bike. In fact we set ourselves the goal of getting close to the times achieved by motorbike star Valentino Rossi, a winner of nine Grand Prix World Championships.

In the short term Yamaha is planning to develop a system of automated tests for its motorbikes to replace those currently carried out by people. Human testing tends to produce variable results, and robotics could certainly help iron out these variations. In the longer term, the humanoid robot could be used for such tasks as grasping tools.

What role does artificial intelligence play in a self-driving car?

When we talk about artificial intelligence we always have to take the context into account, i.e. the environment around the object.

When it comes to self-driving cars, the context – which is basically the environment in which the vehicle has to perform – could be described as pre-determined, given that driving is governed by rules – stopping at a red light, setting off when the light turns green, observing the speed limits in urban areas, etc. In other words, we can teach the software installed in a self-driving car all the rules pertaining to driving. Let me just be clear here: I’m not talking about the thorny dilemma at Google Cars that we heard about recently. The real challenge for artificial intelligence lies rather in understanding the demands coming from human beings. What virtual personal assistants such as Apple’s Siri and Amazon’s Alexa are doing is trying to decipher users’ demands without any prior context – such as for example: ‘What recipe could I make using such and such ingredients?’ This is extremely complex. The fact is that artificial intelligence is far more suited to answering Yes-No questions or carrying out highly specific tasks. So we might say that making a car autonomous is not actually the most complex aspect of the technology.

At the most recent SXSW event, Chris Urmson said that we would not see Google Cars on the roads until they could perform better than cars driven by human beings. Is that the real challenge of the self-driving car?

Probably! If we go back to the question of dilemmas, the most advanced version of the self-driving car would, in a way, avoid coming up against situations which are sensitive from an ethical point of view. The car would be able to anticipate this kind of scenario and would have slowed down for example. But we need more time before we can reach this goal. We’ll probably have to go through an intermediate, semi-autonomous stage first. Although the vehicle will almost entirely take charge of the driving, the ‘driver’ will still be sitting behind the steering wheel in case s/he needs to take back control of the vehicle. But this raises the whole question of how to manage the situation. At what precise moment should the driver ideally take over, and when exactly should s/he be alerted? We could envisage an alarm sounding inside the car, with a countdown system telling the driver that for example “in one minute you’ll be taking control of the car.” It’ll have to be a progressive alert.

robot humanoïde

The humanoid robot created by SRI engineers for the Motorbot project in tandem withYamaha

How soon do you think self-driving cars will be on the roads in California?

Oh, it’ll happen sooner than you might think. Within the next two years, hybrid models will probably be invading our roads. Legislation will then, little by little, need to be framed to support the advent of self-driving cars, by first of all in the next five years authorising the creation of motorway lanes entirely reserved for this type of vehicle. Just imagine a line of autonomous vehicles running in these special lanes, connected to each other and able to learn from each other. A car would then be capable of anticipating the behaviour of the vehicle in front and the one behind. Later on – in ten to fifteen years’ time – we should already see highways mainly set up for self-driving vehicles, with just one lane reserved for older car models. In other words, it’s a combination of progress in both the technology and the law that will lead to the widespread adoption of self-driving cars.

So it’s just a matter of a few years before ‘robot cars’ are out there on our roads. Could we say the same about robots in our homes?

Well, there are already a number of robot solutions designed for the home but they tend to focus on carrying out single, precise tasks. One example is Roomba, a robot-vacuum cleaner made by iRobot, and its competitor Neato. In some ways the most recent washing machines are also a sort of specialised household robot. On the other hand if we’re talking about domestic robots that are able to handle objects smoothly and take on a variety of tasks – tidying a room by picking things up from the floor, clearing the table, and maybe folding the laundry as well – that remains a huge challenge! And actually it’s basically a matter of cost. A robot cannot cost more than $1,000 if it’s to find a real market. But today we can’t yet make one at that price. However, in the next three to four years we should see robots appearing that are capable of manipulating objects in our homes and gradually becoming autonomous!

You’ve developed an autonomous robotic apple picker in Washington State. So we could envisage that in the long term this type of robot deployed in orchards could replace seasonal workers. What’s your view of these developments?

People talk a lot about the impact of robotics on employment. It’s interesting to see that in this case it was a growers’ association that contacted SRI International to develop a robot able to harvest apples, because recruiting seasonal workers was becoming increasingly difficult. The fact is that harvesting apples is a tedious, backbreaking task and the job doesn’t find many takers these days. This example perfectly illustrates the potential for robots to replace manpower on repetitive tasks which very few people nowadays are keen to perform.


Legal mentions © L’Atelier BNP Paribas