The pandemic-induced growth in online shopping, and the shift to same day delivery which was already underway pre-COVID, is resulting in a significant investment in automation of warehouses. This year US$36bn will be invested in warehouse automation globally, up 20 per cent on 2020.
The presence of robots in logistics chains has been growing for some time (Amazon, the world leader, tags its robots with Sesame Street names, such as Kermit the crate-stacker). The ultimate goal is “lights-out warehouses”, where there is no human presence so the lights and ventilation can be turned off. But that is still some time off because robots are only beginning to conquer the challenges of dexterity.
Karen Hao writing in the MIT Technology Review notes that there are two categories of tasks in warehouses: those requiring ‘feet’ to move goods around between staging points in the logistics chain, such as out of storage racks to be stacked beside the conveyer belt for packing; and those requiring ‘hands’, such picking, packing and wrapping into individual shipments to us.
To date, most robotic developments have been in the ‘feet’ tasks, because moving items from point A to point B is fairly simple task (made a little more complex by the need to ‘perceive’ the erratic movements of humans on the shop floor – see the One Thing on smart cranes in Finland). These autonomous mobile robots (AMRs) - like Amazon’s Sesame Street gang – are already commonplace in warehouses.
But robotic arms have only been good at high speed repetition of a simple task, like spot wielding a join on a car body. The ‘holy grail’ of automated logistics is autonomous order picking. Think about when you order a shipment of groceries online – there will be a variety of goods from raw foods, to processed food, to fresh fruit and vegetables and household cleaning products; each item has or is contained within packaging which have very different tactile characteristics, from slippery plastics, tins, paper and cardboard to loose bags; some items are well protected by their containers, such as baked beans, while others are very fragile, such as eggs; and these all need to be selected individually ‘picked’ for each customer order and packed securely and neatly into a delivery box, some wrapped separately in the process.
These are complex ‘hands’ tasks which in most automated warehouses are still done by humans: the robot ‘feet’ bring bulk supplies to the conveyer belts where the ‘human hands’ do the picking and packing. Hence the expression ‘cobots’ or collaborative automation between robots and humans. Some warehouses will use AI to direct the simpler ‘hands’ tasks to an automated process which the AI judges its robot colleagues are up to, and directing everything else to the humans.
Autonomous picking requires sophisticated AI combined with mechanically dextrous robots in order to the achieve the versatility to perform the same range of tasks that human hands are able to do.
A leader in this field is Covariant, in which Temasek and Canada Pension Plan Investment Board have recently taken a major stake. Covariant says its approach is that “instead of learning to master specific tasks separately, Covariant robots learn general abilities such as robust 3D perception, physical affordances of objects, few-shot learning and real-time motion planning. This allows them to adapt to new tasks just like people do — by breaking down complex tasks into simple steps and applying general skills to complete them.” In non-geek, that means the Covariant Brain not only can drive robots across a wide range of logistics streams, including fashion, health and beauty, grocery, pharmaceutical, and general parcel, but also it also draw on the experiences in these verticals so that it can help robots manage items which they have never seen before.
A competitor is Ocado, which started out as an online grocery retailer but soon realised that third party automation products didn’t cut it and so began developing its own AI supply and warehousing products. Its solution to autonomous picking is a swarm of AI controlled droids which ‘scurry’ over a matrix of bins of individual grocery items from which the droids pick to assemble your online order.
What about the humans?
The Financial Times has opined that the coronavirus pandemic (and the UK’s case, Brexit)
“..is changing our perceptions of robots — transformed from enemies that will steal jobs to heroes who can rescue us from chronic labour shortages and boring, repetitive tasks.”
A study based on analysis of the Canadian labour market between 2000-2015 found that “robots are associated with increases in total employment in robot-adopting firms [which] suggest that robots are similar to past generations of general-purpose technologies that ultimately increased labour demand.”
The authors acknowledged their results were at odds with earlier studies which found evidence that AI/robots would be ‘job killers’. The authors offer the possible explanation that while their study looked at the employment impacts when an individual firm adopted AI and robot technology, the earlier studies were based on all firms in a region or sector wide, which may miss expected competitive outcomes between more innovative and less innovative firms:
“robot-adopting firms may experience productivity and employment gains while non-adopting firms in the same industry experience employment and productivity losses. If true, even if robots are observed to cause employment losses at the industry level, it remains unclear whether robots displace workers within robot-adopting firms, or if workers are instead displaced in non-adopting firms due to a decrease in competitiveness.”
However, the study
“surprisingly, did find evidence of displacement of managers, a higher cognitive-skilled occupation that was previously less vulnerable to skill-biased technical change from earlier waves of technology”.
In part, this is because robots’ skills have advanced and they are capable of performing more complex tasks which typically middle management would have performed. However, the larger explanation seems to be that there are fundamental changes in management processes in robot-adopting firms which, in effect, hollowed out middle management.
First, a major task of middle management is to overlook and correct for human error in the manual shop floor processes. If those processes are now performed by robots or a mix or robots and humans, the performance levels should go up, and hence the need for so many middle managers peering over everyone’s shoulders goes down. There was evidence that, following the introduction of robots and AI, the span of control of individual managers over functions and employees increased, and the need for the total number of managers decreased.
Second, and again more surprisingly, there was a reallocation of responsibilities away from middle managers in robot adopting firms – both downwards to the shop floor and upwards to the C-suite. The authors found that “human resource-related decisions with respect to training are decentralized from managers to non-managerial employees, while the choice of production technology is centralized from managers to business owners and corporate headquarters."
There are concerns that as the jobs that can cohabit with robots require low skills pay levels will fall. However, the authors found that “consistent with a decrease in employee monitoring costs, we observe an increase in the adoption of performance pay linked to individual employee performance after robot adoption.” But they did note that with the different skill set mix required by a robot adopting firm, the overall impact on human wages is "ambiguous".
So, while the message is more nuanced than ‘the march of the job killer robots’, the polarisation between low skilled and high skilled jobs with automatization carries significant implications for individual firms and society generally.