Uber beats Waymo to London

PLUS: OpenAI 'will definitely' build humanoids

AuthorAuthorAuthorAuthor

Jennifer Mossalgue, Rowan Cheung, Shubham Sharma & Zach Mink
September 03, 2026

Good morning, robotics enthusiasts. Waymo has been preparing a London launch for months, but Uber and Wayve got there first.

A small fleet of Wayve-powered Mustang Mach-Es are now taking fares in London, running on cameras and radar instead of Waymo-style lidar and HD maps, with a safety driver still behind the wheel.

Also today: Sam Altman says OpenAI “will definitely do a humanoid,” and MIT found a way to understand a robotaxi’s brain.

In today’s robotics rundown:

  • Uber and Wayve bring robotaxis to London

  • Altman says OpenAI ‘will definitely’ build a humanoid

  • MIT’s AI shows what self-driving cars are thinking

  • Skild wants to teach robots with a single video

LATEST DEVELOPMENTS

UBER

Image source: Uber

The Rundown: Uber and UK startup Wayve just launched a robotaxi service in London with supervised self-driving Ford Mustang Mach-Es, getting a head start as Waymo prepares to launch its own driverless service in the city.

The details:

  • The rollout starts with around 20 Wayve-powered Ford Mustang Mach-Es, matched through UberX, Comfort, or Electric at no extra cost.

  • Each car still carries a safety operator, with fully driverless service planned when regulators give the green light.

  • Wayve’s AV2.0 system uses cameras and radar rather than lidar or HD maps, with an AI designed to generalize to new streets, going beyond mapped areas.

  • Waymo is also testing in London ahead of a planned commercial launch, while Lyft and Baidu are preparing rival services.

Why it matters: London is becoming a test bed for two different robotaxi models: Waymo’s sensor-heavy, tightly mapped system and Wayve’s more general-purpose AI approach. Uber is positioning itself as the distribution layer for multiple autonomous-vehicle operators rather than building the driving technology itself.

OPENAI

Image source: Getty Images / Images 2.0

The Rundown: After years of hinting at humanoid ambitions, OpenAI CEO Sam Altman is making it more explicit: He says that OpenAI “will definitely” build a humanoid, alongside specialized machines for data centers and other “form factors.”

The details:

  • Altman told Alex Heath’s Sources podcast that OpenAI “will definitely” build a humanoid, reiterating a comment from Heath’s recent Time interview.

  • Altman said humanoids make sense because the physical world — doors, keyboards, kitchens, machinery — is already designed around the human body.

  • OpenAI is assembling the hardware stack, hiring across custom actuators, mechanical design, simulation, testing, and large-scale robot data collection.

  • The plans could put OpenAI in competition with robot makers it once supplied: former partner Figure dropped OpenAI in 2025 to build its robot AI in-house.

Why it matters: Altman says the robot “brain” matters more than its form factor, but OpenAI is moving beyond software, with infrastructure robots first and perhaps personal robots down the road. A full-stack humanoid would put OpenAI more directly against Figure and Tesla, and could feed it its own stream of real-world training data.

MIT

Image source: MIT

The Rundown: MIT researchers and autonomous-driving company Motional built an AI layer that explains in real time why a self-driving car is making a move, giving humans a better shot at spotting when its reasoning is about to go haywire.

The details:

  • Dubbed CW-Net, it turns a self-driving model’s hidden reasoning into plain-language concepts like “close to cyclist” or “approaching stopped vehicle.”

  • The system was tested in a Motional robotaxi, generating explanations alongside the car’s planned trajectory in real time.

  • In one test, a driver thought the car stopped because it saw a cyclist; CW-Net showed it didn’t see the cyclist, but its automatic emergency braking triggered.

  • Tests with Motional experts and 99 participants found the explanations helped people better predict when the car was likely to behave unexpectedly.

Why it matters: Autonomous cars can be hard to trust because their reasoning is mostly invisible while they’re driving. CW-Net is a step to make that legible, translating a black-box planner into concepts like “approaching stopped vehicle,” so safety drivers and engineers catch a failure forming instead of reconstructing it after the fact.

SKILD AI

Image source: Skild AI

The Rundown: Skild AI unveiled S1, a robot foundation model that takes a single video as its prompt — it can watch a human do a job once, then run the same job on hardware with no fine-tuning and no new training data.

The details:

  • Skild (valued at $14B, backed by SoftBank, Nvidia, and Jeff Bezos) is building a universal “brain” designed to run everything from robot arms to humanoids.

  • S1 handles jobs missing from its pretraining, including potting a plant and flipping pancakes, with some running for 10 minutes across dozens of steps.

  • In the plant test, just 11 minutes passed between starting the human demo and S1 beginning the task autonomously, with no change to the model weights.

  • At 100K hours of pretraining, Skild says that video-prompted S1 hit 66% on unseen tasks against 9% for a language-prompted baseline.

Why it matters: Training a robot on a new long-horizon job can still take tens to hundreds of hours of task-specific data; in Skild’s test, matching S1’s single-video performance took around 380 teleoperated demos, or 50–100 hours. And it’s not alone: Generalist also just unveiled its own one-demo “physical prompting” model.

QUICK HITS

Waymo is launching paid driverless rides in Denver, San Diego, and Tampa, while Amazon’s Zoox is expanding testing to Houston and San Diego.

Tesla is teasing a major Cybercab update in Austin, posting “no steering wheel, no pedals” on X as it prepares to launch its purpose-built two-seat robotaxi.

French robotics startup Wandercraft is seeking €100M ($116M) at a €750M valuation to expand its humanoid business.

Lyte, a robotics startup founded by former Apple engineers, raised $165M at a $1.6B valuation to scale its custom chips and robotic perception systems.

GoPro is being acquired by Starman Holding in a $285M deal that aims to turn the action-camera maker into a broader optics business spanning robotics and AI.

X Square Robot unveiled TwinDEX, a matched wearable exoskeleton and three-finger robot hand that lets humans collect dexterous training data directly on real objects.

Hugging Face’s $399 Microduck sold more than 10K units in days, with CNBC noting that the Pollen Robotics bot runs on a Rockchip processor from China.

TU Delft researchers developed a bird-inspired drone that uses a tactile three-fingered hand to feel for and grip branches, letting it perch to conserve battery power.

Waymo is in talks to raise more than $3B in its first-ever debt financing, tapping lenders Pimco and Blackstone as it bankrolls the expansion of its robotaxi fleet.

KAIST researchers used AI to design a stretchable 3D-printing material, then turned it into a soft robotic hand capable of gripping gently enough to handle fragile eggs.

That's it for today!

Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Zach, Shubham, Jennifer, and Nate — The Rundown’s editorial team