October 6th, 2026Kyle’s Rant
Donna thinks I’m becoming a nutjob.
We had our 30th wedding anniversary the other week and we ended up (as couples do) in a bit of an argument. The thing is, I told her about human brain cells being grown over a computer chip and she said I was a nutjob conspiracy theorist. But surely as a writer with a newspaper, it is my job to write about more than potholes and the inner workings of the Hepburn Shire Council. Anyway, if you are up for something that will blow your mind, read on, and if you’re just a parochial news ingester, move along to her column, Just Sayin’…
Last year a robot ran 100 metres in 21.5 seconds. This year, at the 2026 World Humanoid Robot Games in China, one ran it in 8.64 seconds. For context, the human world record is 9.58 seconds. The robots didn’t just close the gap, they kicked down the door, walked through it and started redecorating.
The numbers from the 2026 Games are the kind that make you pay attention, in the high jump, robots cleared 2.88 metres, shattering the human world record of 2.45 metres. In the 400 metres, one clocked 38.15 seconds against the human benchmark of 43.03 seconds, and this particular machine managed it after tripping mid-race, recovering mid-stride, and finishing anyway. The event itself has gone from 280 teams and roughly 500 robots last year to 666 teams and over 2000 robots from 16 countries in 2026. Whatever’s happening here, it’s not slowing down.
The engineering behind the leap is worth a moment of your time, because the jump from 2025 to 2026 isn’t incremental, it’s a different category of machine. Last year’s robots leaned heavily on cloud computing to make decisions, which introduced small but meaningful delays.
The 2026 models carry onboard neural processing units handling spatial awareness and reflexes in real time. The frames have ditched aluminium and steel in favour of carbon fibre composites and 3D-printed titanium mesh, cutting chassis weight by up to 30 percent. The joint motors have been completely redesigned, and the balance algorithms now fuse LiDAR, high-speed cameras and inertial sensors to predict a stumble before it fully happens and correct for it in milliseconds.
These aren’t just faster robots. They’re categorically different machines from the ones that shuffled awkwardly around a track 12 months ago.
Now, I’ll be honest, we did have a bit of a giggle watching the sprinters hit the finish line. Because here’s the thing – at 14.5 metres per second the current braking systems simply cannot arrest that momentum without doing structural damage to the robot. So the organisers lined the end of the track with heavy crash padding, and one by one these blazing-fast machines sprinted into it at full tilt, caught fire and writhed around in the foam. Spectacular viewing.
Worth noting though that these were stripped of any braking systems, airbags or dragster-style parachutes specifically to squeeze every last kilometre per hour out of them. Fair enough. Engineers have 12 months to solve the stopping problem before 2027, and given what they’ve achieved in the last twelve, I wouldn’t bet against them.
Away from the track, the industrial events are arguably where the real story is. Sorting efficiency is up 400 percent on last year. Robots are successfully plugging complex multi-pin electrical components into mock server racks at a 99.2 per cent success rate, a task that was causing joint-jamming failures in 2025. They’re opening doors, climbing stairs, navigating cluttered rooms and dodging moving people without being prompted. The 2027 mass production target is starting to look less like an aspiration and more like a calendar entry.
And here’s the mind growing (intentional) bit: while all of this is going on in China, a Melbourne startup called Cortical Labs has been quietly doing something that makes the robot story look almost quaint by comparison. Their CL1 device, launched in March 2025, is the world’s first commercial biological computer. Inside a shoebox-sized unit, around 200,000 living human neurons, grown from donated blood stem cells, sit in a nutrient-rich solution on a silicon chip. You deploy code directly to the neurons.
The whole thing runs on about 30 watts of power and keeps the neurons alive for up to six months. This also takes care of the heat issue computers have and the pitch is compelling and bloody unnerving at the same time. You see, silicon chips follow rigid instruction-based patterns but neurons learn, adapt, and rewire themselves based on experience. Combine the two and you potentially have AI systems that don’t just regurgitate existing data but generate genuinely original thinking, at a fraction of the energy cost of traditional silicon computing.
Robots that outrun humans, and a Melbourne mob growing brains on chips -five years ago the stuff of a sci-fi movie.
Yet here we are, rant over…
