Over 20 years ago DARPA (the US government’s defence industry innovation agency) announced a competition to develop autonomous vehicles: the intention was to enable the US military to have a third of its vehicles be autonomous by 2015. Interestingly, the underlying motivation was safety: many US soldiers had been killed and injured by improvised explosive devices in Iraq and Afghanistan and so removing the need for a driver was seen as a way of getting people out of harm’s way: I leave it as an exercise for the reader to work out why the annual death toll of 1.3 million on the world’s roads (of which over 40,000 are in the USA) had not previously been a motivation to do this. Arguably, however, the point at which the race to develop self-driving cars became really visible to the general public was 2009 when Google announced they were starting an autonomous car programme. Since then we have witnessed a perfect case study of the Gartner Group’s “technology hype cycle” as claims for the benefits of autonomy, and how soon it would be available rose to the “peak of inflated expectation”: in 2015 (by which time the US Army had already been disappointed) Tesla claimed that “full autonomy” would be available by 2018. The “trough of disillusionment” follows quickly and in 2021 Oxbotica were saying that full autonomy would arrive in “10 to fifteen years, perhaps never. Today we are probably on the slope of “enlightenment” as, carefully and cautiously, Advanced Driver Assistance Systems (ADAS: examples include automatic lane keeping and speed limiting technologies) and limited domain autonomy are introduced.
But we do well to remember Amara’s law which says that we overestimate the impact of technology in the short term but underestimate it in the long term. Autonomous driving is, in one sense, an application of artificial intelligence and what we have learnt over the last few decades is that the things that humans find hard, such as maths and fast, error-free logical manipulation, are relatively easy to automate. By contrast the things that 3.8 billion years of evolution have made it very easy for humans and other animals to do: use our perception of the world around us to move around the world confidently and safely, are much, much harder to automate than we had imagined. In truth we have had to learn a lot more about how we achieve these things in order to start to build machines that can do them…and there is still a significant gap between the best machine and the average human. But do we really think that that gap will never be closed? The answer is surely “no”, and so the question is not whether this happens, but when: it is taking a lot longer than we thought it would take, but we will deliver these technologies. The interesting question for me is how we will use them to deliver benefits and (spoiler alert) I am not convinced that just replacing taxi-drivers with a robot is the biggest gain for society we can imagine.
Let’s go back that original motivation for the DARPA competition: removing vulnerable humans from dangerous places. Autonomous equipment is already deployed extensively in places like mines and is helping to keep people safely away from dirty and hazardous working environments. Construction, too, is the sort of working environment that has the potential to harm people and, notwithstanding the great progress that has been made by the industry in improving site safety, there are still incidents that hurt and sometimes even kill the people working there. The highways industry is a particular case in point with a good safety record still being punctuated with the occasional incident which disrupts, or even ends, the life of someone who is just trying to earn a living. Why can’t we automate these sites so that people can stay safely away from them (and, ideally, warm and dry as well!) While we are about it the construction industry has had a very poor record of improving productivity: over the last 25 years, whilst productivity in the economy as a whole has grown by about 30%, construction productivity actually declined. Automation has driven dramatic productivity improvements in manufacturing: could it not do the same in construction?
There is more than one reason for why automated machines are not yet widely used in construction. One is intrinsic to the nature of the job: whereas mines are fixed in place, securely fenced off and operate for years or decades which makes it relatively easy to create well-defined spaces in which machines can be used and from which people can be excluded, constructions sites tend to start off as grassy fields, turn into muddy holes and over a few weeks or a few months, into a road or a bridge which makes them loose, ill-defined and ever-changing environments, which it is hard for a machine to understand and easy for a person to wander across.
But there are other reasons too: the construction market is made up of layers of companies, connected by contracts that do not always align incentives well. A highways authority (and there are hundreds of those in the UK) will contract a company to build a road, for example, and that company will, in turn contract others to design and do various parts of the building job. The yellow machines shifting the muck may be owned and operated by a company that is two or three layers of contracts away from the organisation that will end up owning the eventual asset. Productivity considerations can get lost in a series of price negotiations, and coordinating all the various activities on site to make safe and productive use of technology may be much harder when there are people and machines from several different companies, with different objectives, all milling around at the same time.
In short the barriers to the introduction of these technologies are many, and the non-technical considerations are just as powerful as the technological challenges of getting the machines to work properly in the first place.
Given the market structures in construction the most effective way to drive innovation and change is for the client to take the lead: the supply chain is highly competitive, works to low margins and low R&D budgets and the procurement regime makes real innovation unrewarding. So it is to their credit that National Highways, the guardian of the English Strategic Road Network, took the lead to map out how Connected and Automated Plant (CAP) could be introduced.
Picking up the baton TRL and Harper-Adams University are running a workshop on July 17th to build out elements of the roadmap: there is much still to do but progress is being made: how about you come along and help us?
