Robotaxis Are Here: Is Your Next Car Self-Driving? | T-Flash EP10

Today we're announcing Alpamayo 2, an open model for self-driving cars. We're working with car companies across the world. Autonomous driving is one of the most complex engineering problems we have today. It requires developing a safety-critical system that's affordable to be mass-manufactured around the world and has the flexibility to operate in many markets. Welcome to TFLASH. I'm Dennis Yeh, your tech guy. Imagine this driverless scenario on the road. A pedestrian steps out, a truck blocks the lane, and there's no one behind the wheel. A driverless car has to figure it out. Fast. That's the problem NVIDIA is working on. At Computex in Taipei earlier this year, Jensen Huang showed up Alpamayo 2. It's an open AI model that helps self-driving cars think through tricky roads and plan their next move. For Taiwan, the big question is this. Can its chips and electronics power the cars that drive themselves? Today, we're announcing Alpamayo 2, an open model for self-driving cars. We're working with car companies across the world. If you look at these brands that have signed up for the NVIDIA Hyperion, that are building NVIDIA Hyperion cars. This represents about 80% of the world's cars. The manufacturers represent 80% of the world's cars. We are going to have a whole lot of NVIDIA Hyperion systems that are able to run Alpamayo or anybody else's AV stack. We are also connected into mobility services. Approximately 97% of the world's mobility services are connecting with us so that when we deploy Alpamayo on the Hyperion runtime with the Halos operating system we will be able to connect to all of these services across the world. Imagine hailing a taxi and there's no driver. You pick where you're going, the car handles the traffic and the pedestrians on its own. From Tesla's CyberCab in the US to Robotaxi trials in Europe, companies are racing to turn self-driving into a ride service. But can Robotaxis be safe, reliable, and cheap enough for everyday life? Austin, Texas. Tesla's Robotaxi plans are hitting the road. It launched its CyberCab here in October. The goal isn't just a car that drives itself. It's a ride service. No driver needed. But to go big, robo-taxis have to handle messy roads, follow the rules, and convince riders they're safe. The whole invention, development of these cybercabs is humanity changing. I think it'll change transportation from owning your own car. To actually just getting transportation as a service, which is fantastic for disabled people and old people like I'm becoming. Tesla, Zoox and Waymo are all racing to grow driverless rides in the US. But this race is global. In Zagreb, Croatia, local company Vern is testing rides with no driver. It's backed by Uber and runs on tech from China's Pony.ai. Next step? Going from supervised tests to fully driverless. Vern wants to start paid driverless rides in 2027 and expand to more cities in Europe. Six months ago we started the first commercial robot taxi service in Europe. Two months ago we put our vehicles as the first robot taxis on Uber platform and now we're starting the fully driverless robot taxi service for the testing right now without the operator in the vehicle anymore. This is the first one in Europe. But launching is only half the battle. Vern also has to win over riders and make a car with no driver feel safe and comfortable. Feedback is very positive. So we have thousands and thousands of rides and our average rating is 4.8 out of 5. People find it very comfortable, very safe. They feel very relaxed and for the fully driverless rides, it's actually the feedback is like they're a bit boring, which is exactly what we want to be like. So when you feel safe, you feel that it's something very regular, even a little bit boring. London. A car squeezes past parked vehicles with cyclists and people all around. No one's at the wheel. British AI company Wave is testing here. Blind corners, surprise moves, split second calls. Can AI handle London without a human's instincts? Autonomous driving is one of the most complex engineering problems we have today. It requires developing a safety critical system that's affordable to be mass manufactured around the world and has the flexibility to operate in many markets. This is an extraordinarily difficult engineering problem. But the good news is that AI provides a way to develop a level of intelligence that means that we can drive anywhere with mass market hardware. We believe at Wave that there is a new approach that moves away from this mapping, hardware and expense and moves to an end-to-end AI driven approach. This is the same approach that made ChatGPT or AlphaGo possible. For the last 10 years, we've been pioneering end-to-end AI for autonomous driving. And what this means is that we can do away with all of this complexity and replace it with one end-to-end neural network. A kid runs into the road. A cyclist swerves. The car ahead slams on its brakes. Supporters say AI sensors can spot danger and react fast. But can it make the right call every single time? That technology really makes them very safe because the majority of accidents we see are human error. And having that fuller sense of suite that's primarily focused on two things, object detection and classification and path planning and avoidance. So, you know, that makes them very safe. London's black cab drivers know this city inside out. To get licensed, they memorize hundreds of routes and learn to name every street along the way. It can take three to four years. AI reads the road in real time. Humans run on instinct and experience. Can a machine match that and prove it's safe? The core of the knowledge is you learn 320 runs that crisscross away, and then you just repeat. You know you get them in your head and then when you go for exams they'll ask you any random point to any random point in London and you will need to navigate from A to B naming every single street. It is a process and there's nothing like it. It's still really important. Will it be a fad? You know technology is evolving okay we're not we're not riding horse and carts. We're in fact, we're in a £75,000 electric vehicle, so we're always evolving. Are we there yet? No way. I don't think that, I don't want to be like an ignorant taxi driver, like, no, we're not there yet, I just don't, I can't see it. From Austin to Zagreb to London, robo-taxis are getting closer to everyday life. The tech is moving fast. But when something unexpected happens, will AI get it right? Winning riders' trust and proving it's safe might be the hardest part. So how does a car actually drive itself? Before AI can brake, turn, or dodge anything, it has to know what's around it. That means eyes on the road, lots of them. Let's break it down by talking about how a driverless car sees things and moves, and why surprises are still its biggest test. First, the car has to see. Cameras spot objects and lane markings. Radar tracks distance and speed. LiDAR builds a 3D picture of everything around it. GPS and maps tell the car exactly where it is. Put them together, and the car knows where it's going. Seeing is just step one. Next, the car has to make sense of it all. Is that person about to cross? Is the car ahead slowing down? Is it safe to turn? AI crunches the sensor data, tracks what's moving, and spots the risks. Then it decides. Slow down, stop, switch lanes, or wait. Sense, predict, act. Over and over. Nonstop. Here's where it gets confusing. Driver assistance is not the same as driverless. Assistance can steer, manage speed, and keep a safe gap. But you still have to watch the road and be ready to take over. In a driverless car, AI does the driving, but only under certain conditions. So the real question is, does a human still need to keep watch? Once AI makes the call, the car's control systems handle the steering, gas, and brakes. But driving isn't just following a route. The car has to deal with traffic, obstacles, and changing roads, all within the limits of its tech. So not every car with smart features is truly driverless. These hands-off, eyes-off, level three systems, these are going to be in the hands of consumers. And so it's a little unclear. There isn't really much in the way of real regulation that govern these systems today. And so. And so the automakers are... Basically going state by state to get approvals for where they can operate these systems. So how do you tell the difference? Engineers use six levels. Level zero, you do all the driving. Levels one and two, the car helps with steering or speed, but you stay in charge. Level three, the car drives in certain conditions, but you have to be ready to jump in. Level four, it drives itself in set areas, no human needed. Level five, full autonomy. Anywhere a human could drive. Level 3 is the tricky one. These cars are already reaching drivers, but the rules haven't caught up. You know, as we look at level 3 and bringing this to our customers, we need to make sure that they're comfortable and that they trust the vehicle. So where do robo-taxis fit in? Most are level 4. No one behind the wheel, but only in certain areas and conditions. And roads are messy. People jaywalk, it pours, roadworks pop up out of nowhere. So a smooth demo doesn't mean a robotaxi is ready for every road in every kind of weather. Now, imagine a world where cars do the driving for you. Your car changes lanes and navigates traffic while you sit back. No eyes on the road needed. Mercedes-Benz, NVIDIA and General Motors are pushing AI-powered driving forward. The big challenge? Making it safe and earning drivers' trust. NVIDIA and Mercedes-Benz are teaming up to make driving easier with AI. NVIDIA brings the tech. Mercedes builds it into the cars. The system can change lanes and turn at intersections on its own, but the driver still has to be ready to take over. What we have showing today is our L2++ technology. This is basically an address-to-address driving experience that we are delivering to our OEM partners, the car manufacturers, for adopting and delivering the technology to the customer. It's important and it's And it's revolutionary in the sense that it allows the driver to basically I have less burden of the driving, the cognitive load of the driving. The user experience is basically you provide the destination or destinations. You could go to multiple stops and press the start button and the car should take you to your destination. It's quite simple. Mercedes plans to roll this out to more drivers in Germany in the first half of 2027. The goal? Less stress behind the wheel, not replacing the driver. You still need to watch the road and be ready for surprises. It does not negate the need for the driver to be vigilant and supervising the system. Otherwise, the vehicle, the system overall is designed to do all of that. Monitor for close cut-ins, monitor for pedestrians who are jaywalking and suddenly appearing in front of the vehicle. Tire threads suddenly, you know, happen. So the system is still designed to react to all of these situations and instances. But nonetheless, the driver is still required to pay attention at all times. Only now you're starting to see It leaps in terms of improvements and experience that is being delivered by deploying AI on the vehicle. This is why we call this physical AI, the decade for physical AI, where it allows more advanced models being brought into the vehicle, but with the safety harness around it, with the safety architecture that makes sure the AI doesn't do anything that may potentially harm. NVIDIA and Mercedes are using AI to help drivers. GM wants to go further. Tech that lets you take your eyes off the road. Think scrolling your phone or hopping on a video call while the car drives. GM plans to put it in an EV in 2028. But first, it has to prove it can handle the unexpected. So, GM is testing how drivers react in super realistic virtual worlds. So all the research tells us that customers want more and more automated driving solutions, but they want it in a way that is safe and intuitive to use. As General Motors moves into more automated driving scenarios and experiences and solutions around those, what we need to do is make sure that we understand all the different types of scenarios that customers may find themselves in and then understand how customers perceive those scenarios and what they would like to see and do in those scenarios. More freedom means more pressure to get safety right. GM is studying how the tech performs, and how people react when the car does more of the driving. Its Super Cruise system already lets you go hands-free on certain roads, but your eyes have to stay on the road. Going eyes-off? That's a much bigger leap. In today's world with Super Cruise, we deploy a hands-free technology where we still require you to keep your eyes on the road so that we know that you're actively paying attention to every situation. When we talk about eyes off technology at General Motors, what we're talking about is truly a technology that is so safe that it allows you to take your eyes off the road while it's active and do other things in the cabin. Eyes off is when you are able to do another task while you're driving. So you can pull up your phone, you can have that meeting, you can have that conference call. For us right now, we give people candy crush in the simulator because We found it's very effective for getting people involved and wanting to get the highest score. So it's a great way for us to get you engaged and then have to pull you away from a task. So how do you test eyes off driving? With Candy Crush. Inside a fully immersive simulator, researchers watch how people react when the car takes over. They're going to get into the simulator. This all closes. It's fully immersive 360. So you you really feel like you're driving on the road and the vehicle's moving up and down and as you drive down the road you'll get to experience it. It's pretty amazing. We're going to look at your heart rate. Heart rate variability is a great indicator of stress. We want to be able to see, are you comfortable? Is this a convenience? Do you feel like you can trust the system? Trust in our systems is important because then you're going to use it and you're going to use it effectively. If your car can drive while you answer emails or join a meeting, that free time could become something you pay for. GM, Ford, Rivian, and Lucid are all All aiming for eyes off systems around 2028. The money could be big if the tech proves safe and useful. 2028 is kind of the timeline that there's a whole group of automakers that are all targeting for launching their eyes-off systems, including General Motors. Ford is also planning to launch an eyes-off system in that time frame. Rivian, Lucid. Tesla hasn't said when they're really going to go fully eyes-off with their full self-driving system. Going eyes-off is where you... That's really the turning point where consumers will be more willing to pay for this feature because when you go eyes off now. You're really giving, potentially giving the driver back some time because they don't have to be paying attention to the road when they're using the system. With a system like Super Cruise today, they still have to be watching the road, so they can't be doing other things. But when they don't have to watch the road, now they've got time to do other things, like answer their emails while they're commuting to work. And so that's where they see the real value proposition and the potential for consumers to pay that subscription fee. But eyes off leaves almost no room for mistakes. A kid steps out. A car breaks hard. Roadworks force a detour. The car has to handle it all, even when you're not looking. Then there are the rules. In the US, automakers may need approval state by state. That could slow things down. Robot taxis are going from sci-fi to real-world tests, and Taiwan wants in. But going driverless takes more than AI. You need solid sensors, computing power, safe car systems, and clear rules. Kaohsiung is getting ready for a robot taxi service with NVIDIA and Foxconn. Meanwhile, Taipei is testing self-driving buses and preparing its roads. So how close is Taiwan to a driverless future? 全球的汽车产业正从过去的硬体竞赛 A robotaxi carries passengers with no driver, within set limits. It has to see what's around it, guess what others will do, and decide when to turn, slow down, or stop. The hard part? Making safe calls in traffic that doesn't play by the rules. Kaohsiung and Taipei are two very different places to put that to the test. Kaohsiung wants to be one of Taiwan's first robo-taxi cities. In June, NVIDIA expanded its partnership with Foxconn to build level 4 ready fleets on its Drive Hyperion platform. Foxconn plans to launch a service in 2028, first linking the airport to the city and maybe Taiwan's high-speed rail. NVIDIA brings the computing and self-driving tech. Foxconn brings the car design and manufacturing. That could open doors for Taiwanese suppliers in sensors, EVs, and car electronics. But a partnership is just the start. These cars still have to prove they're safe in in real traffic and meet the rules. Taipei? A whole different challenge. Packed streets full of buses, scooters, cars, and people. The city plans to test self-driving shuttles in controlled areas first. Then, a demo zone at Beitou Shilin Technology Park. Then, bus lanes. Four self-driving buses are set for tests at the park and on Xinyi Road's bus lanes in October 2027. The third stage is that in the second half of next year, we will be targeting Taipei's 15 public transport routes to carry out self-driving public transport. Self-driving buses and robo-taxis face a lot of the same problems. Spotting obstacles, reading signals, stopping safely. But buses stick to set routes. Robo-taxis have to handle way more pickups and drop-offs. Taipei's tests could show what roads, traffic systems, and safety plans need to fix before driverless rides can grow. And it's not just about driving safely. Driverless cars also have to protect the software and networks that keep them running. Taiwan is building vehicle safety testing and working with global partners on car cybersecurity. That means checking for weak spots and locking systems down against attacks. There are two aspects to security. How to do it? It must be updated from software, protected by the whole system, protected by hardware and software. And not only this car, from the software design of the entire company, from the whole quality system, to do security processing. We can use the cloud, directly on the cloud, use the OTA method to update the software. Then the owner will, in a formless state, in a state where there is no feeling, drive the car. The more cars run on software and connectivity, the bigger the risk. A cyber attack could steal personal data or even mess with the car's controls. Taiwan's new car cybersecurity rules are expected to kick in in 2028. To win public trust, the whole system has to be protected. The car, the software, the networks, and the emergency plans. So Kaohsiung is gearing up for robo-taxis. Taipei is building experience with self-driving buses and smart roads. Both face the same test. Can driverless cars run safely, reliably, and affordably in everyday traffic? Taiwan's got the tech. Now it has to prove driverless rides can work on its own roads. From Finland to the world, six tiny puppies are winning millions of hearts. A 24-7 livestream has attracted 1.85 million viewers in just two weeks, with fans tuning in from the US and across Europe. Let's take a look. If you liked what you saw, don't forget to search for us on YouTube, give us a thumbs up, and subscribe to our channel. Thanks for watching TFLASH. I'm Dennis Yeh. See you next time.