Robots take over marathons, supermarkets & boxing rings

I'm Nikita. I'm Miraj and you are watching Tech It Out. Nikita, do you think that robots could ever be as athletic as human beings? Well, one robot in South Korea has already completed a marathon, so they're certainly getting closer. Yeah, and it's not just about running either. Robots are stepping into the boxing ring, learning how to handle tasks like restocking supermarket shelves. Which raises an interesting question. Just how far can these machines go well in this episode we'll show you humanoid robots in the boxing ring a four-legged robot that just completed a marathon and humanoids working at supermarkets we'll also show you how to limit the personal data companies collect and share about you all this and more coming up on this episode of Tech It Out I love winters you can go looking for jackets in fact i just got this one but the funny thing is last week i was talking about buying it with a friend and the next thing you know my phone is throwing me all of these ads about where to buy it the same color that i was talking about it's like my phone is listening to me well it's not listening it's actually tracking every click location and purchase is traded across an invisible network of data brokers to understand your spending habits better so much about privacy then well you don't need to be a tech expert to know where and how your data can be shared and how can you opt out of it the kill switches are already built into your devices and operating systems take a look when you visit a website there are two kinds of tracking processes that take place first-party tracking is harmless it remembers your login details or items left in your shopping cart the real concern is third-party tracking add networks embed invisible codes and pixels across thousands of independent websites by connecting those dots with your device unique advertising IDs data brokers compile an intimate dossier of what you read what you buy and where you travel you don't have to accept this surveillance by default in just five minutes you can take control of your digital footprint let's start with Google Chrome click the three dots in the top right corner open settings and navigate to privacy and security click third party cookies and select block third party cookies. That stops external tracking networks from following you across different websites. Now, stop Google from profiling your personal account. Click your profile avatar at the top right, then select manage your Google account. Head to the data and privacy tab on the left and scroll to my ad center. You'll see a switch for personalized ads that's enabled by default. Flip that toggle to off and confirm. Google will stop using your search and watch history to serve tailored ads. If you browse on Microsoft Edge, open settings and go to Privacy, Search and Services. Change your tracking prevention setting from Balanced to Strict. This automatically stops intrusive tracking scripts and ad scripts before they even load. For Apple users, on the iPhone, you head to Settings, tap Apps, select Safari, and turn on Prevent Cross-Site Tracking and Hide IP Address. This stops data brokers from fingerprinting your device. The process is same for Safari. Websites are only half the problem. Mobile apps track you across your phone using hardware advertising IDs. on your phone go to settings open privacy and security tap tracking and toggle off allow apps to request to track on Android head into settings go to security and privacy open ads and tap delete advertising ID once it is deleted apps can no longer stitch your separate mobile activities into a unified profile one last quick tip to protect your data and privacy when websites pop up asking for cookie permissions select necessary only rather than hitting accept all or you can simply click on reject all as well you still see ads online but they won't be powered by your private habits there's something very human very sadistic about putting two blokes in a ring and letting them slug at each other in the Roman times it was the gladiators they were slaves now apparently it's robots at this bout held in Singapore machines aren't thinking for themselves though human operators stand outside the ropes using motion suits to physically guide every blow I don't get it that's just like Nintendo Wii or something isn't it but in real life behind the punches however there is some research going on companies are using the ring to see which parts snap first broken joints delayed signals the tech still has obvious limits take a look at how the bouts played out in Singapore humanoid machines are going head to head in a boxing ring this is the avatar robot combat tournament and instead of autonomous AI calling the shots human fighters are strapped into motion capture suits outside the ropes directing every strike and dodge in real time if the operator slips up the robot takes a hit the tournament is split into two weight divisions in the lightweight class teams field the Unitree G1 it stands 1.32 meters tall and weighs 35 kilos designed mainly for speed and shorter range the heavyweight division introduced the Engine AI T800 to Singapore that bot stands 1.73 meters tall and weighs 75 kilos putting full adult mass and high torque motors on the canvas in a combat scenario actually you can look at it like a stress test so currently in the market you have a lot of humanoid robots right all of them uh are pretty new to people these days like you don't you don't know who's better or what as late in a fight those mechanical forces are delivered at full impact direct strikes and off-balance landings expose how well a biped chassis deals with lateral torque and repeated kinetic shocks however putting humans in the control loop reveals an immediate hardware bottleneck signal delay there's a lag of course between uh the pilot and the robot itself i mean the signal itself has to travel across the distance even though when you see like me or any of the other athletes piloting the robot i punch like this the robot punch like this but there's still a little signal delay like you know half a second but small um and so of course over time i feel like that's going to get hammered out and hammered out more and more audience reaction at the venue showed immense interest in tele-operated sports drawing direct comparisons to early consumer tech adoption cycles I know that this is the start of things you know the robot malfunctioned a bit which is totally fine it's the first of many many robot or humanoid interactions but yeah I'm just kind of scared that this is the start of something either super amazing or things that we'll be seeing in the movies i think this reminds me of how i saw the internet grow in within our society i feel like this is the beginning and i didn't expect much i think today really exceed the expectations that i had and i think it is so exciting katina and the event organizers plan to expand the competition circuit to other international hubs the tech teams argue that data gathered from motor wear latency and impact resistance will directly feed into deploying tele-operated bots for heavy labor hazardous recovery and disaster zones for now the shortest route to stress testing next-gen hardware is putting it in the ring and running it to failure imagine walking into a supermarket and finding robots restocking their shelves in denmark a startup is training humanoids to pick up products handle different packaging and put everything back in its place but you know stocking a shelf isn't as simple as it sounds from slippery pasta packets fragile items these machines must learn the little things that humans can do without thinking the goal is to get these robots working autonomously with the potential to take on more supermarket tasks so how do you teach a robot to work like a human have a look give it an opposable thumb robots are moving beyond factories and production lives taking on tasks that require them to interact with the world around them but working in a human environment means dealing with unpredictable objects changing surroundings and tasks that cannot always be completed the same way twice now a danish startup is training humanoid robots to take on one sad job restocking supermarket shelves at a food store in Lyngby denmark these robots are learning to pick up products handle different packaging and place items where they belong the machines are made by chinese company unitree while hive robots is developing the artificial intelligence that could eventually allow them to do the job autonomously they can learn anything like us humans and the great thing about humanoids is they have legs and feet and arms and fingers so they can walk around in buildings like supermarkets or offices for that matter that are built for humans so you don't have to adapt anything in the stores and we are starting out with replenishing all the different kinds of products on the shelves and we'll move on to trimming or date checking or maybe taking away cardboard boxes or then while we're already walking the aisles we can also start scanning for out of stock products or misplaced items so there's kind of no limits to the very very end goal the aim is to have them restocking shelves through the night while human workers are off the clock for now robots are still learning how to handle the different products that they encounter a supermarket shelf can present a surprisingly complex challenge cans boxes and flexible packets all need to be gripped differently while products can fall or be arranged in new ways from one day to the next a robot that can pick up one item successfully may not automatically know how to handle another that is why Hive robots is training its AI using data collected inside the store rather than relying only on pre-written instructions the process begins with the human operator controlling the robot using virtual reality glasses and two handheld controllers a robot pilot guides the machine through tasks such as picking up cans and restocking shelves these demonstrations generate data that can be used to train the AI model the aim is for the robot to learn from human actions and eventually repeat the same task independently this approach is known as imitation learning instead of programming every movement individually developers use examples of how a task is performed to help the robot learn how to do it for itself but that learning process comes with its own challenges the robot has limited tactile sensing making it harder to judge how firmly it is holding an object there can also be a delay between the operator's movements and the robot's response and some products are more difficult to handle than others we don't want to start with the glass that things that can break are more tricky and for example with pasta as we all know it tends to fall when you place it so if we want the robot to do a good job we need to train it to do the same we do and there are a lot of small things that we as humans don't think about that basically we need to teach the robot pasta packets for example can bend slip or change shape when picked up. Fragile items such as glass jars are not yet a priority. For the robots to restock shelves reliably, it must learn how to handle different objects without damaging them or dropping them. The long-term goal is to move from human-guided training to a robot that can complete these tasks on its own. That means recognizing products, working out how to pick them up, and placing them correctly, even when the shelf for the surrounding environment is different from what it has seen before. And shelf restocking could be just the start. Once the robot can handle products reliably, the same technology could be used to check best before dates, remove empty cardboard boxes, scan for out of stock items and identify products left in the wrong place. Replenishing and filling up a shelf is really the main target at the moment. But we see a huge potential in best before date checks. That's a huge task for our colleagues today. really hard to do and you need to be really detailed and we use a lot of hours in that in that matter every year so our focus is really to to take the replenishment first and then go on to the best before date check each of these tasks could require the robot to apply what it has learned to a different situation the more skills it can master the more work it could potentially take on inside the store five robots hopes to begin autonomously restocking some products at at night by the start of 2027 the company plans to offer this technology as a service allowing retailers to pay for the robots by the hour instead of buying them outright for now the first target is a modest one restocking a limited range of everyday products without human control from there HiFobots wants to improve the machine speed and precision while teaching it to handle more tasks the idea is not simply to build a robot that can move like a human but one that can work spaces designed for people and use the tools and objects already there the Delhi half marathon is on this weekend hats off to all of you who are taking part none of us are we aren't we aren't he likes to sleep on the weekend and i like to play golf now a marathon is tough for most of us In fact, a half marathon is tough enough but imagine a robot running 26.2 miles without a break and without needing a recharge, that's exactly what a four-legged robot in South Korea has achieved RIBO 2 has completed a full marathon on a single battery charge. But this isn't just about a robot that can run, it's about how far robots could go from patrolling streets to reaching disaster zones. A marathon is a test of endurance, 26.2 miles of running where every step takes energy and every minute adds to the fatigue. But what if a robot could cover the same distance without stopping to recharge? In South Korea, a four-legged robot has done just that. Meet Drybot 2. The quadruped robot completed a full marathon in 4 hours and 19 minutes, maintaining an average speed of 2.64 meters per second. But the real achievement was not just how fast it moved. It was that it finished the entire race on a single battery charge. For mobile robots, that is a significant challenge. quadruped robots can walk across uneven terrain and navigate complex environments but their power consumption can limit how far they travel before they need to recharge that can make long outdoor missions difficult to sustain, quadruped robots have barely been usable outdoors at all, they were mainly used indoors patrolling small fixed areas along a set course, our robot by contrast can walk up to 50 kilometers which means it's capable of outdoor operations over very long distances RIBO 2 can travel up to 50 kilometers on a single charge opening up the possibility of operations over much longer distances, so how does it use its energy more efficiently? The answer lies in how it learned to walk, the research team used a robot physics simulation platform called RiboSim developed by Professor Jemin Wangbo in 2018, instead of training the robot only through physical trials the team simulated its movements inside a virtual environment, there the robot could experience different walking patterns and terrain conditions without having to perform every movement in the real world, these simulations helped train an AI based walking control system designed to reduce unnecessary movement and wasted energy the system also works to limit the impact when the robot's feet hit the ground as well as slipping an excessive joint movement each of these factors can affect how much power a robot uses to move by learning more efficient ways to walk Ribo 2 can cover more ground without draining its battery as quickly and that could matter far beyond a marathon RiSim can be thought of as a dynamic engine what we do is mathematically model the robot system and simulate it so simulation accuracy matters and assuming everything is accurate speed becomes especially important as well the reason speed matters is that we collect data based on simulation and use it for training." Ribo 2's developers say they have received interest from South Korea's National Fire Agency, the military and industry. Disaster response is another potential application, where robots may need to cover long distances while searching areas that are dangerous for human rescuers. Longer battery life could allow them to spend more time working and less time returning to recharge. The team also sees potential in regular street patrols. A robot that can move for longer periods without interruptions could be more practical for routine monitoring and inspection. But completing a marathon is only one measure of a robot's ability. Real-world operations can involve unpredictable weather, obstacles, different surfaces and tasks that require more than walking. The challenge will be to show that the same energy efficiency can be maintained outside controlled tests and across a wider range of conditions. Still, the Ribo 2's marathon offers a glimpse of what more efficient robot movement could make possible. The goal is not simply to build a robot that can walk, it is to build one that can keep going. And when a robot can cover more distance on a single charge, its usefulness may extend well beyond the finish line. world AI and humanoids learning to navigate the real world and now getting intelligence it's really beginning to look like science fiction you know well maybe but there's a difference between doing a task and understanding what it means to be human I mean sure but that's the same argument we use when we said that they can't be sentient and now we have all of these AI agents showing decidedly human behavior and characteristics I really don't know anymore Nikita and that's something worth thinking about on that note that's it from us this week have a great weekend keep watching Beyond World as One see you next week bye bye bye