My first fight with a robot started badly.
I punched it hard in the chest and felt a jarring sensation running back up my arm. It was like hitting a brick wall.
Things got better when I started kicking it.
I gave it a good shove with my foot and it went reeling back into the ropes of the ring.
The robot stumbled back after it was kicked and eventually lost its balance
At this point, a human might have given up, but robots are indefatigable. It bounced straight back up, ready for further punishment.
So I kicked it again.
Reader, I am not proud of this, but in my defence, it was the robot or me. I couldn’t let humanity down.
To clarify, the robot was fine with all this. Well, its owners were anyway – I didn’t exchange many words with my opponent, mainly because it couldn’t talk.
Robot didn’t stand a fighting chance
Chinese robotics company Unitree invited me to fight one of its G1 units live on stage at the Consumer Electronics Show (CES) in Las Vegas, the world’s biggest technology fair.
The event was arranged as a demonstration of the robot’s ability to work alongside humans. It had no chance of beating me or even landing a blow.
The robot stumbled into the ropes several times
Once I’d recovered from the bout, I was able to appreciate the strengths and weaknesses of the technological achievement I’d just witnessed.
Balance is tricky for robots, but the G1 was able to get back up even after a hefty kick.
What it lacked was agility. It wasn’t able to dodge my blows the way a human might.
In part, this was by design – Unitree wants the humans to enjoy the experience, which means letting them win.
Manthorpe triumphed over the robot
But at the same time, this reveals a basic technological challenge faced by any humanoid robot manufacturer: making a robot nimble and light on its feet means equipping it to deal with unexpected instability.
Humans don’t just balance – we rethink the task mid-motion. For the most part, robots are still figuring that out.
The question of balance could stand for the field of robotics as a whole.
Robot butler still a long way off
I came to CES to uncover the truth about robots.
The hype around physical AI is almost overwhelming, and not without good reason – the success of self-driving cars shows that machines can be taught to perform as well, if not better, than the human equivalents.
Nvidia boss Jensen Huang summed up the mood when I asked him about robots earlier this week.
We will have robots with human capabilities “this year”, he said, “because I know how fast the technology is moving”.
But, as I found, the fantasy of the robot butler rushing around the house to do all the chores is still a long way off, simply because human bodies and behaviours are so hard to reproduce.
Laundry-folding robot
One of the most impressive technical achievements I saw during my visit to CES was a robot folding laundry.
The awareness and dexterity needed to manipulate soft materials have always been beyond robots, until now.
However, when I investigated the laundry-folding robot further, I found that it required four days of remote operation just to get it used to the new table and lights of the unfamiliar CES setting.
Dyna, the company behind it, rents it to hotels at $3,000 to $5,000 a month – and you can see the appeal because the robot can fold for 16 hours a day. Why not 24? Because it can’t take the laundry away. It hasn’t learnt how to do that yet.
The robot can fold towels, but not take them away
Wherever you turn, you see both how far robots have come and how far they have to go.
Over the last few days, I have seen dozens of robot hands, trying to replicate all the different facets of the incredible tool we are gifted at birth.
Touch, for instance, is something technologists do their best to replicate by planting tactile sensors on the robot’s fingers.
Or the way that human fingers have a bit of give to them, which allows them to absorb impact rather than fighting against it.
Tactile sensors in these robotic fingertips can measure the weight of the object the robot is trying to grab
In robots, the ability to yield rather than lock up is called “back-driveability”, and is one of the cherished goals of the industry.
A humanish-feeling handshake
South Korean company WiRobotics showed me their new actuator – the robotic term for muscle – which has in-built force recognition designed to create back-driveability.
“It can absorb the impact and even utilise the impact,” says Dr Yong-Jae Kim, founder and CEO of WiRobotics.
His robot achieved a rare feat: a humanish-feeling handshake. That’s extremely impressive from a technical point of view, but it should give you an indication of how far away we are from robot Jeeves.
Even if robots do achieve perfect back-drivability, that’s just the beginning.
The handshake of this robot felt almost human-like
Human hands aren’t always yielding. We constantly switch between softness and rigidity – gripping a phone gently, then bracing to lift a suitcase.
That dynamic switching is the real goal. And at this stage, it still feels some way off.
Read more from Sky News:
Changes to Musk’s AI bot ‘insulting’, No 10 says
The 40 jobs ‘most at risk’ from AI – and 40 it can’t touch
What’s changed is that now, thanks to generative AI, there’s a path to get there. Every roboticist I spoke to was in agreement: generative AI has revolutionised the industry.
“Robotics has completely changed. We can make robots feel more like people. We can make them do things we’ve never been able to do before,” says Aya Durbin, humanoid application product lead at Boston Dynamics.
It does seem now as if humanoid robots are going to get here. It might just take a bit longer than many people hope and fear.
