logo
How college students built the fastest Rubik's Cube-solving robot yet

How college students built the fastest Rubik's Cube-solving robot yet

The Verge2 days ago

A team of Purdue University students recently set a new Guinness World Record with their custom robot that solved a Rubik's Cube in just 0.103 seconds. That was about a third of the time it took the previous record-setting bot. But the new record wasn't achieved by simply building a robot that moves faster. The students used a combination of high-speed but low-res camera systems, a cube customized for improved strength, and a special solving technique popular among human speed cubers.
The Rubik's Cube-solving robot arms race kicked off in 2014, when a robot called Cubestormer 3 built with Lego Mindstorms parts and a Samsung Galaxy S4 solved the iconic puzzle in 3.253 seconds — faster than any human or robot could at the time. (The current world record for a human solving a Rubik's Cube belongs to Xuanyi Geng, who did it in just 3.05 seconds.) Over the course of a decade, engineers managed to reduce that record to just hundreds of milliseconds.
Last May, engineers at Mitsubishi Electric in Japan claimed the world record with a robot that solved a cube in 0.305 seconds. The record stood for almost a year before the team from Purdue's Elmore Family School of Electrical and Computer Engineering — Junpei Ota, Aden Hurd, Matthew Patrohay, and Alex Berta — shattered it. Their robot has come to be known as Purdubik's Cube. Bringing the robot record down to less than half a second required moving away from Lego and, instead, using optimized components like industrial motors. Getting it down to just 0.103 seconds, however, required the team from Purdue to find multiple new ways to shave off milliseconds.
'Each robot that previous world record-holders has done has kind of focused on one new thing,' Patrohay tells The Verge. When MIT grad students broke the record in 2018, they opted for industrial hardware that outperformed what previous record-holders had used. Mitsubishi Electric chose electric motors that were better suited for the specific task of spinning each side of the cube, instead of just hardware that moved faster.
However, the first thing the Purdue students improved was actually the speed that their robot could visualize the scrambled cube. Human speed cubing competitors are allowed to study a Rubik's Cube before their timer starts, but the robot record includes the time it takes it to determine the location of all the colored squares. The students used a pair of high-speed machine vision cameras from Flir, with a resolution of just 720x540 pixels, pointed at opposing corners of the cube. Each camera can see three sides simultaneously during exposures that lasted as little as 10 microseconds.
Although it may seem instantaneous, it takes time for a camera to process the data coming from a sensor and turn it into a digital picture. The Purdubik's Cube uses a custom image detection system that skips image processing altogether. It also only focuses on a very small area of what each camera's sensor sees — a cropped region that's just 128x124 pixels in size — to reduce the amount of data being moved around.
Raw data from the sensors is sent straight to a high-speed color detection system that uses the RGB measurements from even smaller sample areas on each square to determine their color faster than other approaches — even AI.
'It's sometimes slightly less reliable,' Patrohay admits, 'but even if it's 90 percent consistent, that's good enough as long as it's fast. We really want that speed.'
Despite a lot of the hardware on Purdue's robot being custom-made, the team chose to go with existing software when it came to figuring out the fastest way to solve a scrambled cube. They used Elias Frantar's Rob-Twophase, which is a cube-solving algorithm that takes into account the unique capabilities of robots, like being able to spin two sides of a cube simultaneously.
The team also took advantage of a Rubik's Cube-solving technique called corner cutting where you can start to turn one side of the cube before you've finished turning another side that's perpendicular to it. The advantage to this technique is that you're not waiting for one side to completely finish its rotation before starting another. For a brief moment, there's overlap between the movements of the two sides that can result in a significant amount of time saved when you're chasing a world record.
The challenge with corner cutting is that if you use too much force (like a robot is capable of) and don't time things perfectly, you can physically break or even completely destroy a Rubik's Cube. In addition to perfecting the timing of the robot's movements and the acceleration of its motors, the students had to customize the cube itself.
Guinness World Records follows the guidelines of the World Cube Association, which has a long list of regulations that need to be followed before a record will be recognized. It allows competitors to modify their cube, so long as it twists and turns like a standard Rubik's Cube and has nine colored squares on each of its six sides, with each side a different color. Materials other than plastic can be used, but the color parts all need to have the same texture.
To improve its durability, the Purdue team upgraded the internal structure of their cubes with a custom 3D-printed version made from stronger SLS nylon plastic. The WCA also allows the use of lubricants to help make cubes spin more freely, but here it's used for a different reason.
'The cube we use for the record is tensioned incredibly tight, like almost hilariously tight,' says Patrohay. 'The one that we modified is very difficult to turn. Not impossible, but you can't turn it with your fingers. You have to really get your wrist into it.' When solving the cube at high speeds, the lubricant helps to smooth out its movements while the increased tension reduces overturns and improves control so time-saving tricks like corner cutting can be used.
Faster servo motors do help to reduce solving times, but it's not as simple as maxing out their speed and hoping for the best. The Purdubik's Cube uses six motors attached to metal shafts that slot into the center of each side of the cube. After testing several different approaches the team settled on a trapezoidal motion profile where the servos accelerate at speeds of up to 12,000,000 degrees/s2, but decelerate much slower, closer to 3,000,000 degrees/s2, so the robot can more accurately position each side as it comes to a stop.
Could the Purdubik's Cube break the record again? Patrohay believes it's possible, but it would need a stronger cube made out of something other than plastic. 'If you were to make a completely application-specific Rubik's Cube out of some sort of carbon fiber composite, then I could imagine you being able to survive at higher speeds, and just being able to survive at higher speeds would then allow you to bring the time down.'

Orange background

Try Our AI Features

Explore what Daily8 AI can do for you:

Comments

No comments yet...

Related Articles

Kepler K2 "Bumblebee" Debuts at ICRA 2025, Captivating Attendees
Kepler K2 "Bumblebee" Debuts at ICRA 2025, Captivating Attendees

Yahoo

time30 minutes ago

  • Yahoo

Kepler K2 "Bumblebee" Debuts at ICRA 2025, Captivating Attendees

ATLANTA, June 3, 2025 /PRNewswire/ -- During the IEEE International Conference on Robotics and Automation 2025 (ICRA 2025) held in Atlanta from May 19 to 23, Shanghai Kepler Robotics Co., Ltd ("Kepler Humanoid Robot") introduced its latest industrial-grade humanoid robot, the K2 "Bumblebee," drawing global attention with a live demonstration of its real-world operational capabilities. The event brought together leading researchers, executives, and engineers from across the robotics ecosystem. Notable visitors to the Kepler Humanoid Robot booth included Jim Fan, Senior Research Scientist at NVIDIA, and Hesheng Wang, General Chair of the International Conference on Intelligent Robots and Systems (IROS), among other robotics experts. They engaged in discussions on the trajectory of humanoid robotics and the path to scalable deployment in industrial environments. Kepler's booth quickly became a high-traffic destination at ICRA 2025. The K2 "Bumblebee" greeted attendees with natural gestures, navigated the venue with steady, autonomous mobility, and interacted fluidly with other robotic systems. Its engaging presence prompted a flurry of photos and videos, with many attendees lining up to pose with the robot. Debo Hu, CEO of Kepler Humanoid Robot, stated, "The next major focus for the humanoid robot sector is achieving a complete commercial value loop, and industrial environments present the clearest path to near-term deployment. Our fifth-generation humanoid robots are now in limited-series production, with industrial use cases as our initial focus. Our latest round of funding reinforces our confidence in the commercial viability of our platform." Kepler Humanoid Robot is dedicated to developing industry-leading blue-collar humanoid systems. Purpose-built for industrial settings, the K2 "Bumblebee," stands 175 cm tall, weighs 75 kg , and features 52 degrees of freedom and more than 80 integrated sensors. Its onboard compute reaches 100 TOPS, enabling it to autonomously execute task sequences in structured operational environments. As the humanoid category moves into early-stage mass production, robots must demonstrate high payload capacity, endurance, and cost-efficiency to achieve sustainable deployment. The K2 "Bumblebee" features Kepler's proprietary planetary roller screw actuators, engineered for smooth motion control and reduced energy draw. With near-zero static power consumption, the robot offers extended uptime. Combined with its innovative series-parallel architecture and advanced actuation system, the K2 supports payloads of 15 kg per arm and up to 30 kg with both arms. The robot delivers up to 8 hours of operational time on a single 1-hour charge. Additionally, its proprietary rotary actuators enable millimeter-level precision for fine motor tasks. In addition, the Kepler K2 "Bumblebee" also includes Kepler's proprietary Dexterous Hands, each featuring 11 degrees of freedom, 25 force-sensing contact points per finger, and a 6-axis force/torque sensor at the wrist. These capabilities allow for advanced manipulation tasks across a range of industrial applications. Kepler reports that over 80% of the robot's core hardware is developed and manufactured in-house—supporting vertical integration, reducing costs, and improving supply chain resilience. The base model of the K2 "Bumblebee" is priced at $30,000 and can perform the equivalent workload of approximately 1.5 full-time human workers in comparable timeframes. Deployment flexibility is a key enabler for humanoid adoption. According to Goldman Sachs, humanoid robots are expected to begin integration into manufacturing environments between 2024 and 2027. Kepler is targeting intelligent manufacturing, warehousing and logistics, niche industries, academic research, and education as its primary markets. The company's commercialization roadmap prioritizes penetration of specific verticals before expanding into general-purpose use cases. K2 "Bumblebee" units are currently being trialed across a range of real-world industrial scenarios. Kepler continues to expand its partnerships throughout the robotics value chain to accelerate the transition from prototyping to commercial-scale deployment. View original content to download multimedia: SOURCE Shanghai Kepler Robot Co., Ltd. Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data

OpenAI's Altman sees 2026 as a turning point for AI in business
OpenAI's Altman sees 2026 as a turning point for AI in business

Yahoo

time33 minutes ago

  • Yahoo

OpenAI's Altman sees 2026 as a turning point for AI in business

STORY: :: June 2, 2025 :: San Francisco, California :: Sam Altman says 2026 will be a big year for AI solving problems and making discoveries :: Sam Altman, CEO, OpenAI 'I think we'll be at the point next year where you can not only use the system to sort of automate some business processes or fill these new products and services, but you can really say, I have this hugely important problem in my business. I will throw tons of compute at it if you can solve it. And the models will be able to go figure out things that teams of people on their own can't do." 'I would bet next year that in some limited cases, at least in some small ways, we start to see agents that can help us discover new knowledge or can figure out solutions to business problems that are kind of very nontrivial. Right now, it's very much in the category of, okay, if you got something like repetitive cognitive work, we can automate it at a kind of a low level on a short time horizon." 'So what an enterprise will be able to do, we talked about this a little bit, but just like give it your hardest problem if you're a chip design company, say go design me a better chip than I could have possibly had before. If you're a biotech company trying to cure some disease, so just go work on this for me. Like that's not so far away.' Speaking alongside Conviction founder Sarah Guo and Snowflake CEO Sridhar Ramaswamy, Altman said companies prepared to harness the full potential of AI will experience a 'step change' as models evolve from automating routine tasks to tackling non-trivial challenges. 'I would bet next year that, at least in some small ways, we start to see agents that can help us discover new knowledge,' Altman said, adding that future systems may significantly accelerate scientific discovery. Se produjo un error al recuperar la información Inicia sesión para acceder a tu portafolio Se produjo un error al recuperar la información Se produjo un error al recuperar la información Se produjo un error al recuperar la información Se produjo un error al recuperar la información

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into the world of global news and events? Download our app today from your preferred app store and start exploring.
app-storeplay-store