A robotaxi can look like an ordinary car with extra equipment on its roof. A passenger sees a familiar booking button. Somebody still has to keep the vehicle ready for its next trip and meet the requirements of the city where it operates.
That work interests me in Pony AI's partnership with Uber. Their British announcements put Chinese self-driving technology alongside a service passengers already use. They also leave room for a local business to run the vehicles.
London and Manchester are testing a service
On October 8, the partners announced plans to test Pony AI's latest robotaxis in London. Six days later, they announced Manchester with their Croatian mobility partner, where vehicle testing was already underway. These announcements give us no start date for a public service without a safety driver.
The European partnership already divides responsibilities in Zagreb. Pony AI supplies the self-driving technology. The Croatian partner handles the local fleet, while Uber connects the service with riders. The Manchester announcement brings those partners into another market.
I find that division of work more interesting than another clip of a car driving itself. It gives a transport operator something specific to consider: which part of putting these vehicles to work could its existing business provide?
The customer relationship may sit with a different company from the one that builds the driving system.
That changes how I would approach a potential partnership. A conversation about operating vehicles has different requirements from a conversation about supplying driving software. The familiar app gives the technology a route to passengers, but the fleet still needs an organisation that can keep the service running locally.
Factory trials need somebody responsible for safety
A similar question about the work around a robot appears in South Korea's approval of a humanoid trial. The plan covers 20 robots at four industrial sites, with proposed jobs including handling car parts and work in chilled logistics. Each site needs its own risk assessment.
Picture a humanoid carrying a part along an aisle. A person can step into its path. The load changes how it balances and how much harm a collision could cause. A safety assessment has to address that situation.
A walking machine sharing a factory floor creates a different problem from an arm operating inside a fenced area. The trial offers a supervised way to gather evidence in those workplaces. Approval lets that learning begin; operating results still have to follow.
For a manufacturer, there is work here for the people who design the process around the machine. The proposed task needs to be specific enough for an engineering team to assess what happens when people and moving loads share the route.
The engineering handover is part of development
Unitree's founder has described another practical difficulty inside robotics development. Engineers read papers and write software, train a model, and pass it to people testing the machine. Results return to the engineers for another attempt.
His proposed direction is for AI to take on more of that development work and use test results to improve the next version. People would still judge whether the robot's performance makes sense for the job. A completed movement can be useless in its intended setting.
I would look at the handovers in an engineering team's own testing process before discussing how this might help. The remarks describe a direction for development, with no firm date for broad factory adoption.
Kawasaki's collaboration with Japan's Matsuo Institute gives a specific example of training a movement. The team trained a humanoid in simulation before transferring the movement to a real machine, which lifted a loaded basket onto a table. The physical demonstration took place in San Jose in May; the training work was announced in October.
The basket leaving the floor is the useful moment to examine. The robot has to balance as the weight moves through its arms. An industrial buyer can ask how the training would change for its own loads and working conditions.
Follow the cart through your building
Japan's Kachaka Evo makes a smaller automation task easy to picture. The compact mobile robot drives underneath a cart or shelf, raises it and carries it elsewhere. Its rated carrying capacity is 100 kg. The cart has to fit the machine, and the route has to suit the job.
I like being able to count those journeys and measure the time they take. That gives an operator something to compare with the cost of automation. Before choosing a robot, follow a load through your building and identify the repeated journey you would want a machine to take over.
[00:00] Pony AI is bringing its Robo-taxi technology to London and Manchester with [music] Uber. You may soon see Chinese self-driving technology behind a very familiar-looking button. Getting a car onto a London street is only part of the challenge. The partnership tells us something about how Chinese robotics companies can reach customers overseas. South Korea is opening a path for humanoids to work beside people in factories. That puts the next stage of testing inside workplaces with real safety responsibilities. And Unitree's founder wants AI to take over more of the work of developing robot software. The people teaching the machines may get some automation of their own. Small Japanese robot can pick up the cards a
[00:49] factory already owns. That could make a transport upgrade much easier to fit into an [music] existing operation. Kawasaki has a new example of teaching a humanoid physical task in a virtual world. The interesting moment comes when the real machines has to lift the load. >> [music] >> And at the end, I have a little cart-moving robot that deserves its own parking space. Every week, I connect these developments to decisions you can make in your business. I live in Shanghai, I speak Chinese, and I visit the companies building these machines [music] every week. Reading the original sources helps me bring you the detail behind the headline. If you would like to see the technology with your [music] team, our private robotic tours are linked in the description.
[01:37] >> Asia base >> Now to the first story. Pony AI is taking its Robo-taxi partnership to two British cities. Pony AI [music] is a Chinese company that develops self-driving technology. Uber is a booking app many passengers already know. On October 8th, the partners announced plans to test Pony AI's latest Robo-taxis in London. Six days later, Pony AI, Uber, and Vern announced Manchester, [music] where vehicle testing was already underway. The announcement is about testing. It gives us no start date for a public service without a safety [music] driver. For a passenger, a robot taxi can look like a car with some extra equipment on the roof. For the company running it, the driving software is only part of the operations. Someone has to own the vehicles and keep them ready for
[02:25] the next trip. The service has also to meet local requirements. Pony AI already has a European partnership with Uber and Vern, a Croatian mobility company. In Zagreb, the partners have divided up the work. Pony AI provides the self-driving technology. Vern handles the local fleet, and Uber connects the service with riders. The Manchester announcement brings the same three partners into another market. Pony AI supplies the driving technology, Uber brings access to riders, and Vern contributes the service operation. I find the partnership more interesting than another video of a car driving itself. [music] A Chinese technology company can reach a new market through a service people already use. Local partners can build a business around putting that technology to work. If you run transport
[03:15] services, look at which part of that operation you could provide. The customer relationship may sit with a different company from the one [music] that builds the driving system. Would you book a robot taxi through Uber if there were still a safety driver in the car? Tell me in the comments what would make you comfortable trying it. Now to South Korea. South Korea is making room for humanoid factory trials. South Korea has approved a regulatory trial involving humanoids from Roborus, a Korean robot maker. The plan covers 20 robots at four industrial sites. The proposed jobs include handling car parts and work in chilled logistics. The approval let the project test these machines in workplaces shared with people. Each site needs its own risk assessment. The approval is a step toward running the trials with operating
[04:03] [music] evidence still to come. Think about a robot carrying a part down an aisle. People can step into its route, the load changes how the robot balances, and how much harm a collision could cause. The safety assessment has to deal with that whole situation. A fixed robot arm can work inside a fenced area. [music] A walking machine sharing a factory floor creates a different safety problem. South Korea is creating a supervised way to gather the evidence needed for that problem. For a manufacturer, this opens a route to testing [music] tasks that have been difficult to automate. It also creates work for the people who design safe [music] processes around the machine. Permission to test is where that learning starts. If you know someone planning a robot trial, send them this
[04:51] episode. This is super interesting for them. And now to Unitree. Unitree wants AI to help develop robot software. We have covered Unitree's factory ambitions before. The new detail from its founder, Wang Xing Xing, concerns the work inside the engineering [music] team. At Shanghai's industry fair, he described a development process with plenty of manual handovers. Engineers read papers and [music] write the software. They train a model and pass it to people who test the robot. The results go back to the engineers for another round of changes. Wang wants AI to take on more of the coding, training, and testing process. [music] The idea is to let the system use test results to improve the next version. He also said people still need to judge whether the robot is doing something useful. [music] A robot can complete a movement and still do it in a way that
[05:39] makes no sense for the job. A person has to decide what good performance means in that setting. His proposal points to another way robotics should develop faster. Better tools [music] could shorten the work between an unsuccessful test and the next attempt. His remarks describe a direction for development with no firm date for broad factory adoption. [music] If you manage an engineering team, look at the handovers in your own testing process. The feedback loop is part of the product. And now to Japan. Japan's Kachaka [music] Evo can move the carts a factory already owns. Preferred Robotics, a Japanese robot maker, demonstrated Kachaka Evo at the CEATEC Technology Fair. The company opened sales on October 1st with shipments scheduled to begin during the month. It
[06:26] is a compact mobile robot that drives under a cart or shelf. It raises the load and carries it to another position. It's rated carrying capacity is 100 kg. That design starts with something a factory already has. You can picture a worker pushing the same cart along the same route throughout the shift. The robot takes over the movement by lifting the cart from underneath. The company [music] says the machines maps its surroundings with magnetic tape on the floor. That gives a site more flexibility when the layout changes. The cart still needs to fit the machine and the route still needs to suit the job. I like this approach because the task is easy to understand. You can count the journey and measure how much time they take. Then you have something useful to compare with the cost of automation. Before choosing a robot, follow a load
[07:15] through your own building. You may find a repeated transport job that's a small machine can take over. For more of my observations from China, follow me on LinkedIn through the link in the description. Kawasaki is teaching a humanoid to lift through simulation. We looked at Kawasaki's Kaleido humanoid in the previous episode. This update is about a new training collaboration with Japan's Matsuo Institute. [music] Matsuo Institute published results from a demonstration in which Caledo 9 lifted a loaded basket and placed it on a table. The physical demonstration took place in San Jose in May. The training work was announced in October. Matsuo says they trained the movement in a computer simulation before transferring it to the real robot. A simulator lets the team
[08:04] try movements without putting the physical machine through every attempt. The real lift then test how well that learning carries over. Picture the basket leaving the floor. The robot has to keep its balance as the weight [music] moves through its arms. Contact with the handles and small difference in the load can change what happens. For an industrial buyer, this is a useful example of a way to develop robot skills. It gives you a specific task to discuss with the supplier. You can ask [music] how the training would change for your own loads and working conditions. Kawasaki brings the physical machine to this collaboration. The AI team brings methods for teaching in a new movement. That division of work could give manufacturers another route to adding capabilities. Across these stories, the opportunity extends into
[08:51] the work around the robot. The local operator and the engineering team can each have a business to build. If you want to work out where your company fits, come and see this technology in China. Use the robotics tool link in the description and tell me which process you want to improve. We can discuss a visit built around your industry and the questions you need answered. This little robot has earned its parking space. Before you go, watch Kachaka Evo Duck underneath the cart. It picks it up and takes it away. I suspect the person who usually pushes that cart is paying [music] very close attention.
