Image: © jim/Stock.adobe.com
Elon Musk says a humanoid Tesla bot is the company’s ‘most important product development’, but what are the challenges with bringing this sort of tech to being? Sam Cox asked Irish robotics expert Conor McGinn.
Whether it’s the clearly mechanical C-3PO or the skin-covered T-800 Terminator, the concept of modelling robots in our image has been seen in sci-fi for years. But including two arms and two legs is a design decision – while humans have evolved like this, it doesn’t mean it’s necessary for robots.
As AI advances and technologies improve, some scientists do think crafting humanoid robots is the next logical step. Indeed, demonstrations have been coming out in all shapes and sizes, with Boston Dynamics’ Atlas robot being able to show off impressive parkour skills.
But many were not impressed with Elon Musk’s Tesla Bot demonstration last year. A dancer in a body suit came on stage in what Musk described as the model for a humanoid robot the company would produce in future, but what The Verge described as “a distraction and an empty promise”.
Musk said the robot would incorporate Tesla’s autotech to help humans with mundane and repetitive tasks, and on an earnings call last week claimed it was the company’s “most important product development”. But questions have been raised about how this could actually be achieved.
Dr Conor McGinn is CEO of growing Irish start-up Akara, a Trinity College Dublin spin-out developing AI tech and robots for the healthcare sector. His team started out with Stevie, a social care robot designed to interact with older people, before moving on to Violet, a robot that can autonomously navigate a room and disinfect it using ultraviolet light.
From his experience, McGinn believes there are three particular problems when it comes to designing bipedal, humanoid robots.
The missing link
“The first is that a lot of the theory still isn’t there. Very often, when it comes to designing humanoid robots, we’re reverse engineering,” says McGinn.
“So we’re taking our understanding of nature, and we’re trying to build, and there are still large gaps in that knowledge. So that means that the challenge is much more than, ‘Can we build it?’ If you don’t have the blueprint for what you’re making, then it becomes even harder.”
The problem at the heart of this is that many things that are easy for humans aren’t always so easy for robots. While a computer can calculate sums in a fraction of a second, AI can struggle to achieve simple tasks that require human understanding.
“So if, for example, we have a computer vision algorithm that is able to recognise apples on the table, saying, ‘Is there an apple on the table?’, the machine might be able to say yes or no.
“But if you want a machine to understand whether this is an apple that can be safe to eat, or whether it’s a cooking apple, or whether it’s a poisoned apple, then that requires a much more detailed understanding.”
There is a worldwide effort to bride this gap. McGinn brings up the example of RoboCup, an annual international competition with the aim of advancing intelligent robotics.
“Within the competition, they have certain categories,” he explains. “And one of them, for example, is soccer. So people build soccer-playing robots; the idea being that if you make a robot play soccer, those skills will transfer into more useful domains.”
It’s not just sport in the spotlight. The competition also has a category that focuses on AI robots that can carry out simple tasks such as shopping.
“Even in those competitions, where the complexity of the environments is reduced, we see that the level of performance is nowhere near as good as what a person would do,” McGinn explains. The reverse engineering required for these tasks is monumental, and often of questionable value.
“In a setting like a shopping centre, where shopping trolleys are used, the need for [a robot with] legs would be…I would question it.
“I personally would advocate for building more accessible environments where wheeled robots can be used, rather than designing very expensive and complicated machines to navigate poorly designed and inaccessible locations.”
The rules of the game are different
The second point McGinn highlights is that for humanoid robots to become a sight on our streets, designers need to account for complex modern environments and recognise that some aspects of the human world are much more complicated than others.
“We say the rules of the game are different,” he says.
The decision-making process “is not very complicated” with something like roads, where there are defined structures, directions, rules. “Generally, getting from A to B is a problem that we can solve. We have a map of the environment and roads don’t change on an overnight basis,” McGinn adds. “On the other hand, human-centred environments are very different.”
If you have robots walking down footpaths, around buildings and other spaces where humans tend to move all around the place, there aren’t the same types of rules to follow. “It’s not going to be controlled for in the same way roads are, where cars drive one behind the other. People clutter and crowd.”
‘If we build something that looks very humanlike and it doesn’t perform very humanlike, that’s going to draw people’s defences pretty quickly’
– CONOR MCGINN
The physical environment isn’t the only challenge these robots would have to overcome. McGinn notes that if you build a machine that looks like a person, then people will likely develop expectations in line with that appearance. He recalls the old expression: “If it walks like a duck and quacks like a duck, it is a duck.”
“If we build something that looks very humanlike and it doesn’t perform very humanlike, that’s going to draw people’s defences pretty quickly,” he adds. “You know, the evidence would indicate that people will get very frustrated with technology very quickly.”
McGinn says that in studies of human interaction, it has been seen that people don’t mind being flexible when collaborating with another person if the experience is positive. He gives the analogy of an exchange student in your home – they might not be fluent in your language and you may find it difficult to communicate with them. But they’ll smile or do something endearing, something human, and your frustration with the situation subsides.
Phones, on the other hand, don’t have this effect. If Google Maps refuses to open or Facebook Messenger keeps crashing, you’ll likely get irritated. So, finding a way for people to adopt robots is just as important as being able to create them.
The sheer scale
The third and final problem that McGinn highlights is the sheer scale and complexity of building a humanoid, bipedal robot. This is why designing this type of robot is not just difficult, but will take time.
The process requires “input from many, many people with many different expertise”, from design to physical manufacturing to complex software and updates.
“There’s infrastructure that needs to get built and, in our experience, there’s probably as much code written off the robot to facilitate those kind of operations as there is on the robot. Logistically, a lot of consideration needs to be given to how many people will work on something like this in a coherent way.
“If you’re dealing with a bespoke bipedal robot that involves, you know, arm manipulation and grasping, there’s just a ton of custom software that’s going to need to get built. And most of this is sequential, you can’t really do it in parallel.”
‘You need to figure out what the robot’s doing and design its complexity in response to the complexity of the environment it needs to operate in’
– CONOR MCGINN
Overall, McGinn believes the discourse around robot design is important. He points to AI winters – the term given to periods where funding for AI development is reduced – and says these were often the result of hyped expectations and overestimations.
Both the media and the AI community have a responsibly to temper expectations, while remaining ambitious. An integral part of this is knowing what to create next – working smarter rather than harder. While humanoid robots may have a place in a future society, that doesn’t mean they’re the inevitable next step.
“The way humans have evolved over thousands of years, we’ve evolved with bodies we can’t dynamically change,” McGinn explains.
“It’s designed inherently as an all-rounder. It’s got a huge amount of redundancy built into it, because we live 80 to 100 years old and over the course of those years we need our body to do an awful lot for us. We also developed a lot of redundant features, a lot of things we don’t need very often, so I would strongly question if you were building a robot why you would overengineer it like that.
“Would it not make more sense to design core parts of it? For example, in a bipedal robot, let’s suppose you’ve realised you definitely do need to have walking. Are you going to design it to have 10 toes, because balance is improved by 5pc if you have 10 toes? Probably not.
“So when it comes down to it, you need to figure out what the robot’s doing and design its complexity in response to the complexity of the environment it needs to operate in.”
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
