➡ China’s new robot, G1, is a 132cm tall humanoid that can walk, run, jump, and even play soccer with human-like movements. It uses advanced technology to process its surroundings and can operate for two hours on a single charge. The G1 comes in four different models, each with varying features and capabilities. China’s Pudong district has launched a robot base to train these robots, aiming to have 1000 by 2027 to enhance manufacturing and service applications.

China’s new robot base plans to train thousands of humanoids as Unitary releases a shocking AI agility update. But what can it do and how much will it cost? To start, the 132-centimetre-tall G1 robot weighs 35kg and incorporates cutting-edge features like 3D Lidar and a real-sense depth camera to process environmental data. As a result of this, plus a range of new data, the G1 robot has rapidly improved its walking, balance and running to be faster and more natural looking than possibly any other humanoid robot available for purchase so far. And this is partially thanks to the robot’s redirection algorithm, which combines interactive mesh with inverse kinematics technology while factoring in things like optimization, the robot’s end-pose constraints, joint positioning and velocity limitations.

As a result of this, the G1 robot has already set a record with a 1.4-metre standing long jump that demonstrates its agility using up to 43 degrees of freedom and 120Nm of torque in its joints. Plus, the G1 outperformed in other areas by executing tasks like walking, jogging, climbing stairs, dancing and even playing soccer with remarkable speed and human-like motion. And for its brains, the G1 includes eight high-performance CPU cores connected to WiFi 6 and Bluetooth 5.2. This power allows the robot to reach speeds of up to two metres per second or seven kilometres per hour and can operate for up to two hours on a single charge using its 9000mAh battery.

Additionally, the G1’s joints are especially impressive in this update, with three degrees of freedom in the knees and shoulders, and two in the elbows, wrists and ankles, with this flexibility allowing the robot to execute intricate manoeuvres including splits and folding into a compact shape for easier transportation. But even more impressive is the robot’s ability to recover even when walking on uneven or slippery surfaces while still maintaining an extremely natural human-like gait. And when it comes to cost, the G1 is selling at four different price points, with the standard model serving as the baseline to offer essential features for basic tasks, whereas the Plus model upgrades to NVIDIA’s Jetson Orange chip with 100 tera operations per second for better programming and AI performance.

Then, there’s the Smart model which enhances mobility by increasing hip freedom from 1 to 3 degrees and arm freedom from 5 to 7 degrees to enable more complex movements. And finally, the top-tier Ultimate model includes all mobility upgrades and adds two DEX31 force-controlled, three-finger dextrous hands for precision tasks. However, while these hands lack multi-touch sensors, a haptic feedback upgrade is available. Additionally, with Unitary quickly outpacing the development speed of other industry leaders like Tesla and Boston Dynamics, the G1 robot is quickly becoming a dominant player in the humanoid robot market.

And the G1 is able to endure falls as well as recover from physical impacts like kicks and punches, proving it’s built to survive tough real-world conditions. In fact, this ruggedness is even supported by its latest AI upgrades, allowing the robot to learn through reinforcement and imitation methods, with new updates and abilities delivered remotely via over-the-air downloads regularly. But which model do you think is best, and how much would you be willing to pay at the current level of performance? This is the question on everyone’s mind, as China’s Pudong district just launched a full-blown humanoid robot base that’s currently training 100 robots and has plans to expand to 1000 by the year 2027 in order to boost Chinese manufacturing and service applications.

More importantly, the center will partner with local robotics companies to collect a data set of 10 million high-quality physical data entries, with this initiative seeking to improve humanoid robots’ effectiveness in the real world to prepare for mass rollouts. For example, as China faces an aging population and growing global tech competition, humanoid robots are viewed as solutions to ongoing workforce shortages with the country’s humanoid robot market being projected to reach $11 billion by 2030. And on top of this, the Pudong center also plans to debut its next-generation humanoid robot called Deep Snake, which is designed with advanced flexibility and intelligence.

Additionally, Beijing will host the first world humanoid robot sports games later this year. And as production scales up, experts predict China could sell humanoid robots at prices between $15 to $30,000 by 2030, which would undercut most competitors and make advanced robotics more accessible worldwide. And in another agentic AI breakthrough, Google just shipped a series of upgrades to its Gemini AI assistant to near-general intelligence, focusing on multimedia capabilities, app integration, user accessibility and more. But the main feature that has everyone talking is currently Gemini Live, which enables users to interact with images, files, and YouTube videos directly within their conversations.

Additionally, users can ask Gemini to analyze photos and provide insights, though this functionality is currently exclusive to Samsung Galaxy S24 and S25 phones, or Google Pixel 9 devices, with Google promising to extend these features to more devices. And looking ahead, Google also plans to launch Project Astra, to adding screen-sharing and live video streaming to Gemini. Plus, Gemini’s functionality now extends to more apps through extensions, meaning that while it’s already linked to platforms like YouTube, Google Maps, Gmail, and Spotify, it can now also control various Samsung apps on the Galaxy S25, such as Calendar, Notes, Reminder, and Clock.

Another notable new feature also allows users to combine extensions within a single command. For instance, Gemini can find recipes and save them to Samsung Notes or Google Keep in one step. This functionality is available across web, Android, and iOS platforms, with Galaxy S25 users soon being able to replace Samsung’s Bixby Assistant. And for Gemini Advanced users, the app will also introduce a deep research feature aimed at providing more in-depth research capabilities. And finally, OpenAI just unveiled Operator as their newest AI assistant that’s capable of independently navigating the web. But at the core of Operator is the computer-using agent, which is a breakthrough AI model that enables the assistant to process website content as raw data, combining GPT 4.0’s vision capabilities with advanced reasoning.

In fact, Operator doesn’t even require special integration with websites, but instead uses screenshots to view web pages, allowing it to tap, click, scroll, and even fill out forms or order groceries on its own. Users can even customize their interactions by adding personalized instructions for specific sites or general use. Operator also supports multitasking, running different browser tasks in separate chat windows, making it a powerful tool for handling routine online activities seamlessly. Importantly, Operator uses a three-phase process, capturing screen content, using chain-of-thought reasoning to decide actions, and executing them, with its inner-monologue approach minimizing errors for more accurate results, and testing shows it excels on real-world websites, achieving an 87% success rate on tasks like navigating Amazon or Google Maps.

However, it struggles with complex interfaces such as combining PDFs or managing calendars, with a lower success rate of 38.1% in advanced benchmarks. And for security, Operator includes three protective layers by requesting user permission for critical actions like payments, allowing users to monitor sensitive activity in watch mode, and detecting and halting suspicious behavior with a dedicated model. [tr:trw].