China's inaugural 'Robot Olympics' delivers impressive feats and disastrous falls

The World Humanoid Robot Games pitted androids against each other in Olympic-style sporting events.

The first-ever World Humanoid Robot Games have come to a close with some new world records, but don't expect them to beat humans in a 100-meter dash any time soon. The three-day robotics event in Beijing, China that saw humanoid robots compete in everything from boxing to cleaning concluded this weekend. According to the World Humanoid Robot Games, more than 280 teams from 16 countries, including the US, Germany, Brazil and the host country, entered their robots into the event.

A majority of the teams came from universities, but several teams came from established robotics companies like Unitree and Fournier Intelligence. Hundreds of robots competed in traditional sporting events like running, soccer and table tennis, along with events for more practical tasks, like cleaning or sorting medicine. Unitree snagged gold medals in four categories, including the 1,500-meter, 400-meter, 100-meter, and 4x100-meter dash events. According to Unitree, the fastest speed achieved by one of its robots during the event was 4.78 m/s, which is roughly 10 mph.

In the 100-meter sprint, Unitree boasted a 33.71-second time, but that's nowhere near the world record holder Usain Bolt's time of 9.58 seconds, which was achieved in 2009. Instead, most teams took this event as an opportunity to showcase the abilities of their designs, while also stress-testing their robots in a competitive environment. We're still a long way from widespread adoption of these humanoid bots though, especially if they can't avoid running into each other.

Read More

 

Agility's Digit warehouse robot understands natural language commands thanks to AI smarts

“Take the box that’s the color of Darth Vader’s lightsaber, and move it to the tallest tower in the front row.”

Agility Robotics shared a demo video Wednesday of one of its Digit robots upgraded with AI. Although that may conjure terrifying pop-culture images of sentient sci-fi machines taking over the world, the demo video reveals something much more pedestrian, if not boring. The bipedal warehouse robot ploddingly works to complete a slightly puzzling task without direct human control or detailed guidance. In the clip, it slowly but successfully interprets and executes the command, “Take the box that’s the color of Darth Vader’s lightsaber, and move it to the tallest tower in the front row.”

The company, which added a “head” and “hands” to Digit earlier this year, pitches the demonstration as a glimpse into how large language models (LLMs) can enhance its humanoid machines. It suggests it’s a natural fit, describing Digit as “a physical embodiment of artificial intelligence.”

“We created a demo space with a series of numbered towers of several heights, as well as three boxes with multiple defining characteristics,” Pras Velagapudi, Agility Robotics’ vice president of innovation and chief architect, explained in the clip. “Digit was given information about this environment, but was not given any specific information about its tasks. Just natural language commands of varying complexity to see if it could execute them.”

The clip shows Digit’s beat-by-beat inner monologue overlaid on the screen as the robot thinks its way through the task. For example, “I should find out what commands I can use,” “I should first assess the environment to see what items and locations are available,” and “Darth Vader’s light saber is red, and so the red box (box2) is the one I need to move. The tallest tower in the front row is tower4. I will first move to the location of box2.”

After moving to the location of the red box, Digit continues. “Now that I’m at the location of box2, I should pick it up,” it reasons. It continues, “Now that I have box2, I should move to the location of the tallest tower in the front row, which is tower4” and “Now that I’m at tower4, I should place box2 on it.” Finally, it declares its work finished: “I believe I have completed the task, as I have moved the box that is the color of Darth Vader’s light saber (red, box2) to the tallest tower in the front row (tower4).”

Agility Robotics, which pitches Digit as sparing human workers from the most taxing physical activities, says it created the demo to “show how LLMs could make our robots more versatile and faster to deploy.” The company is building an Oregon factory to produce 10,000 humanoid robots annually. It has also inked a deal with Amazon for the retailer to test Digit in a Seattle-area facility. Fiction-fueled fears aside, the robots are much more likely to hurt humans by stealing their warehouse jobs than by shapeshifting, murdering innocents or reenacting other Hollywood-fueled dystopian nightmares.

Video link

Read More

 

Apptronik’s Apollo is the latest humanoid robot to beat Tesla to market

The 5-foot-8, 160-lb Apollo can lift up to 55 lbs.

Apptronik unveiled a new workforce robot today. Named Apollo, the machine is designed to “work in environments designed for, and directly alongside, humans.” The android is initially intended to move and carry cases and totes in logistics and manufacturing settings. But the Austin-based Apptronik sees Apollo expanding into “construction, oil and gas, electronics production, retail, home delivery, elder care” and more. Apollo follows Xiaomi’s reveal of the CyberOne robot last year, which looked remarkably similar to the still-unreleased Tesla Bot.

The 5-foot-8, 160-lb Apollo can lift up to 55 lbs. (Apptronik says it optimized efficiency by making its arms lighter than the weight they can lift.) It uses swappable batteries — running up to four hours per pack — which should provide more flexibility than robots that require wall charging before springing back into action. “In short, this battery-based approach means greater work output for Apollo and greater operational efficiency for customers,” Apptronik wrote in a press release today.

Apptronik views Apollo as a robot that can adapt to the job. The company says it built “modularity into Apollo’s design, empowering users to decide whether Apollo is best used for their applications as a true bi-pedal walking humanoid, a torso that operates on wheels or one mounted in a stationary location.” The robot has digital panels on its face and chest to provide a “friendly, human-like countenance” to make workers feel comfortable working alongside it (as it potentially moves towards automating their jobs).

Apptronik hasn’t announced public pricing for the robot. You can read more about Apollo on the company’s product page.

Read More

 

This vacuum robot dog can find and suck up trash with its feet

VERO uses AI to pick up cigarette butts and small bits of trash off of Italy’s beaches and other public spaces.

Cigarette butts pose a huge risk to the world’s oceans and can be a pain to clean up by hand especially on public spaces like beaches. A group of Italian scientists have built a quadruped robot that can identify litter and pick up the smaller bits with its leg mounted vacuums.

VERO, the vacuum equipped quadruped robot, is a four-legged device designed to look for and clean up litter on a variety of terrains. VERO was designed and built by a team of researchers from the Dynamic Legged Systems lab at the Italian Institute of Technology in Genoa, according to USA Today.

The group published a paper back in April on VERO’s development and effectiveness in the Journal of Field Robotics. The research paper states that cigarette butts are a serious concern. Discarded butts release toxic chemicals and microplastics into the ocean as they break down. It’s also the “second most common undisposed waste worldwide, in terrains that are hard to reach for wheeled and tracked robots.”

VERO is designed for picking up this common type of small litter. An operator sets up a field target for the robot to traverse. Then it slowly walks the entire length of the target while identifying litter with a special neural network and onboard cameras. The quadruped robot has a “convolutional neural network for litter detection” that can target litter and pick it up with one of four leg mounted vacuums, according to IEEE Spectrum.

Cleaning up beaches also can be a challenge because the sand makes it hard to lug wheeled trash bins or heavy receptacles over the terrain. The researchers conducted tests on “six different outdoor” scenarios to show VERO’s proficiency at navigating difficult terrain. It can steady itself while picking up trash with an Intel RealSense depth camera mounted on its chin.

The robot didn’t get every piece of trash in its initial test but it still picked up 90 percent of the cigarette butts identified in testing. That’s 90 percent less waste that ends up in the ocean.

There don’t seem to be any plans to implement VERO just yet. The researchers say VERO’s design could be programmed and engineered to do other tasks like spraying crops, looking for weaknesses in infrastructure and helping with construction projects.

Read More

 

These robots move through the magic of mushrooms

The fungus-powered machines could lead to future crop-tending robots that sense subtle chemical reactions.

Researchers at Cornell University tapped into fungal mycelia to power a pair of proof-of-concept robots. Mycelia, the underground fungal network that can sprout mushrooms as its above-ground fruit, can sense light and chemical reactions and communicate through electrical signals. This makes it a novel component in hybrid robotics that could someday detect crop conditions otherwise invisible to humans.

The Cornell researchers created two robots: a soft, spider-like one and a four-wheeled buggy. The researchers used mycelia’s light-sensing abilities to control the machines using ultraviolet light. The project required experts in mycology (the study of fungi), neurobiology, mechanical engineering, electronics and signal processing.

“If you think about a synthetic system — let’s say, any passive sensor — we just use it for one purpose,” lead author Anand Mishra said. “But living systems respond to touch, they respond to light, they respond to heat, they respond to even some unknowns, like signals. That’s why we think, OK, if you wanted to build future robots, how can they work in an unexpected environment? We can leverage these living systems, and any unknown input comes in, the robot will respond to that.”

The fungal robot uses an electrical interface that (after blocking out interference from vibrations and electromagnetic signals) records and processes the mycelia’s electrophysical activity in real time. A controller, mimicking a portion of animals' central nervous systems, acted as “a kind of neural circuit.” The team designed the controller to read the fungi’s raw electrical signal, process it and translate it into digital controls. These were then sent to the machine’s actuators.

The pair of shroom-bots successfully completed three experiments, including walking and rolling in response to the mycelia’s signals and changing their gaits in response to UV light. The researchers also successfully overrode the mycelia’s signals to control the robots manually, a crucial component if later versions were to be deployed in the wild.

As for where this technology goes, it could spawn more advanced versions that tap into mycelia’s ability to sense chemical reactions. “In this case we used light as the input, but in the future it will be chemical,” according to Rob Shepherd, Cornell mechanical and aerospace engineering professor and the paper’s senior author. The researchers believe this could lead to future robots that sense soil chemistry in crops, deciding when to add more fertilizer, “perhaps mitigating downstream effects of agriculture like harmful algal blooms,” Shepherd said.

You can read the team’s research paper at Science Robotics and find out more about the project from the Cornell Chronicle.

Read More

 

Avride’s next-gen delivery robot ditches two wheels and adds NVIDIA AI brains

Autonomous delivery vehicle company Avride has a fresh design — and NVIDIA AI brains. The company’s engineers have swapped out the old six-wheel configuration for a more efficient four-wheel chassis. It can make 180-degree turns almost instantly, effortlessly park on inclines and move faster without compromising safety.

Avride has been working on autonomous delivery robots since 2019. It began as part of Russian tech company Yandex’s autonomous driving wing. But the spun-off company divested its Russian assets after Vladimir Putin ordered the invasion of Ukraine in 2022 and rebranded as Avride. It’s now owned by the Netherlands-based Nebius Group (formerly Yandex N.V.), headquartered in Austin, TX and making deals with the likes of Uber.

The company’s latest delivery robot shakes up one of the few constants from previous iterations: They all had six wheels. The new four-wheel robo-buggy uses a “groundbreaking chassis design” that eliminates some of the rough spots from older generations. These included additional friction and tire wear caused by excessive braking required for turns, lower maneuverability and less precise trajectory execution. Avride says the new model dramatically improves on all of those counts.

The new vehicle’s wheels are mounted on movable arms attached to a pivoting axle. For turns, each wheel glides along a circular path stabilized by the central arm. “This design allows the wheels to rotate both inward and outward, reducing friction during turns,” the company wrote in its announcement blog post.

Central to the new design is ditching the traditional front and rear axles for mechanically connected wheel pairs on each side. Avride says this enables simultaneous turning angle adjustment, leading to more precise positioning and maneuvers.

Among the results of the fresh approach are almost instant 180-degree turns. Avride says this especially helps when navigating narrow sidewalks, where sudden adjustments could be necessary. Parking on slopes is also more energy efficient: It now sets its wheels in a cross pattern to park in place without careening downward. The tighter controls also let the company increase its maximum speed. “This means faster deliveries for our customers,” the company wrote. (And, presumably, more profit.)

Not only did the new generation of delivery bots get a new body, but it also got smarter. Powered by the NVIDIA Jetson Orin platform, essentially an “AI brain for robots,” the vehicles can now tap into neural networks as powerful as those in full-size autonomous cars. This lets them process “vast amounts” of sensor data like lidar inputs and camera feeds in real time.

Finally, it wouldn’t be a delivery buggy without a cargo compartment — and that got an upgrade, too. The new model has a fully detachable storage section, allowing for modular swap-outs for different purposes. Avride says its standard cargo hold is big enough to hold several large pizzas and drinks or multiple grocery bags. It also adds a sliding lid that only provides access to the correct section, helping to avoid delivering orders to the wrong customers.

Engineering and design nerds can read much more detail about the new robots in Avride’s Medium post.

Read More