Entertaining as it is to watch robots summersault, it’s important that we remain realistic about what robots really can and can’t do. The real world is bound by safety and product regulations – and of course by the question of whether the robot is economically viable for a company to adopt. Regulatory and commercial decisions need to be grounded in a clear understanding of what’s actually possible today, and a recognition that most technology takes between 10 and 25 years to be adopted at scale.
In commercially-available robotic applications, AI is making most impact in expanding robot mobility and dexterity. Until recently, robots in factories and warehouses have moved along pre-determined routes, guided by signals (magnetic, laser, lidar) from devices installed for this purpose in their environment. Sensors enable these robots to recognise an obstacle and stop to avoid collision, continuing on their route when their path is clear. However, these robots cannot find an alternative route to their goal if the obstacle remains in their path. In contrast, an AI-enabled mobile robot gets from A to B by building a real-time map (or updating a pre-programmed map in real-time) of its environment and of its location within that environment, planning a path to the programmed goal, sensing obstacles and re-planning a path in-situ.
Meanwhile, advances in 3D vision technology - one of the AI technologies making fastest progress – mean that robots can now identify objects even when they are partly hidden by other objects or poorly lit. Machine learning enables the robot to teach itself in a very short time how to pick up an object it has not encountered before, applying the appropriate level of force. The machine learning algorithm continues to improve as it picks.
The combination of intelligent mobility and vision technology is driving a boom in automated guided vehicles (AGVs) and picking robots in warehouses and factories. AI-enabled autonomous vehicles are already at work in factories and warehouses, checking inventories, fetching goods and picking items from bins.
The IFR projects a five-fold increase in service robots in logistics by 2020. However, most picking robots designed for agricultural use are still not economically viable compared to human labour. It is still very difficult for robots to pick objects that have irregular and variable shapes – such as fruit and vegetables – or objects that are not rigid – for example, goods in plastic wrapping. These challenges are being overcome, but what’s less clear is how long it will take for broad-scale adoption.
Another rapidly-advancing area of AI is natural-language processing (NLP). This has wide application in software ‘chatbots’ that can understand and respond to questions and provide information. In robots, NLP combines with mobility to enable mobile information robots that assist customers in environments such as hotels, hospitals, airports and shops. They can answer questions, lead customers to requested products or locations and video-link the customer to a human service agent.
Considerable research is going into applying AI to collaborative robots that are designed to work with humans in factories, hospitals and warehouses to assemble products, lift patients and package goods. AI is not yet widely applied in collaborative robots because enabling the robot to respond on the fly to the situation it encounters brings in a level of unpredictability that is unacceptable in most industrial and commercial settings for safety and quality reasons. We’re unlikely to see robots solely powered by AI algorithms in widescale use on the shop floor anytime soon. What we’re more likely to see is the application of AI algorithms to help robots learn to perform tasks quickly. Once the task has been learned, all or a large part of it can be hard-coded to ensure predictability. This is very promising for small-to-medium sized businesses that work with short production runs as it enables them to set up the robot quickly and train it on a new task as a new customer or production run comes on board. AI can also be used for continuous optimisation of the robot, analysing its motion and recommending changes that, though they may be small, can have large cumulative impact on the robot’s speed, efficiency and energy use. AI can also be used to signal when a robot is about to need maintenance, saving companies the expense of machine down-time.
As AI continues its rapid trajectory, important questions are being raised about how its use should be governed by companies and governments. Specific regulation regarding the use of AI in commercially available robots doesn’t yet exist, and the IFR believes that this would not be productive. Robots are physical machines and as such are already governed by detailed safety standards and regulations - such as the EU Machine Directive and OSHA Guidelines for Robotic Safety in the U.S, and various ISO standards. It makes most sense to incrementally adapt these existing guidelines to reflect developments in robots in specific contexts.
The number of AI start-ups in the US topped 600 in 2016 and 128 robot start-up companies in the US received new funding in the same year. There is no doubt that we are at the start of a phase of rapid expansion in robotics, much of it driven by advances in AI. Software development requires little investment in fixed capital, and in many cases can be rolled out to users with bugs that are fixed on the fly. Robots, on the other hand, require fixed capital investment and are governed by safety standards that mean they must be bug-free and fully predictable when made commercially available. So while we can enjoy the rush of box-office films with humanoid, sentient robots, we won’t be seeing them in factories, warehouses and hospitals anytime soon.
Picture © Shutterstock