➡ Foundation Robotics Labs has unveiled the world’s first robot soldier, Phantom MK-1, which can perform various tasks including military service, industrial use, and more. The robot is designed to handle tasks requiring strength and precision, and can adapt to different environments. In other advancements, NVIDIA and Carnegie Mellon University have developed a new framework, ASAP, that improves the performance of AI-powered robots, and PND Botics has showcased a robot, Adam, capable of performing intricate tasks with human-like precision. Lastly, OpenAI and Google have introduced AI tools that can autonomously conduct multi-step research on the internet.

This is the world’s first robot soldier. As San Francisco-based startup Foundation Robotics Labs just revealed its newest Phantom MK-1 humanoid, which is engineered for various general-purpose tasks, including going to war. In terms of its size, the Phantom stands 175 centimeters tall and weighs 80 kilograms, with the robot’s frame being designed to handle a wide range of general tasks. But these actuators deliver the performance of hydraulics while maintaining the efficiency and quiet operation of electric motors. These actuators enable Phantom to handle tasks requiring both strength and precision, with a maximum peak torque of 160 Newton meters and a back driving torque of less than a single Newton meter, allowing the robot to lift a 20 kilogram payload as a result.

With this strength, the company says it’s focusing on a wide range of general purpose applications including military service, industrial use, entertainment, home service, and more as the robot continues learning. And when it comes to operations, Phantom’s upper torso serves as its all-in-one hub, integrating high resolution cameras, advanced sensors, compute units, and the robot’s battery. This design equips the robot to process complex information, perceive its surroundings, and execute tasks with the highest degree of accuracy, with the sensors and cameras providing Phantom with 360 degree situational awareness for it to adapt to dynamic environments and interact with objects or people more effectively.

Additionally, Phantom’s modular design allows for easy upgrades to ensure that the robot always stays up-to-date no matter what add-ons come out in the future. Then there’s the robot’s lower torso, which also serves as Phantom’s foundation for mobility and stability. Equipped with advanced actuators and a sophisticated balance system, the robot can navigate stairs, traverse uneven terrain, and move through confined spaces, with Phantom’s human light gate allowing it to operate in environments designed for humans, making it adaptable to offices, warehouses, and disaster zones alike. And to demonstrate this, the robot’s ability to maintain balance in unstable conditions or while carrying objects further, proves its suitability for real-world tasks requiring both dexterity and strength.

But NVIDIA and Carnegie Mellon University just made an even greater breakthrough in AI-powered robotics with the development of their brand new framework called Aligning Simulation and Real Physics, or ASAP for short, which finally closes the gap between simulated training and real-world performance by significantly reducing errors between virtual and physical movements, and it works in two stages. First, robots are trained in a simulated environment, where they can safely and efficiently learn complex movements. Then, ASAP applies a specialized model that accounts for real-world differences and learns to detect and adjust for variations between the virtual and physical domains.

As a result, ASAP reduces overall motion errors by approximately 53% when compared to existing methods, with this level of precision marking a significant step toward enabling robots to perform complex tasks reliably outside of controlled environments. And testing it in the real world on a G1 humanoid led to impressive results, with the robot executing challenging maneuvers such as forward jumps spanning over one meter, as well as various movements of several world-famous athletes. But despite its success, the project also highlighted some existing hardware limitations, as the dynamic nature of these movements caused motors to overheat in some cases, resulting in two robots being damaged during testing.

However, the researchers remain optimistic, viewing these challenges as opportunities for refinement and the team made their code publicly available on GitHub, encouraging other researchers to build upon their work. And in yet another breakthrough that used the G1 robot, researchers have introduced a new humanoid motion framework that allows robots to sit, roll, and stabilize themselves using all body parts, rather than being restricted to traditional bipedal movement. This advancement addresses long-standing challenges in model predictive control and reinforcement learning for robots, where unpredictable contact sequences often make real-time motion planning nearly impossible.

But the real key to this breakthrough lies in its GPU accelerated rigid body simulator, which allows for the rapid training of a whole body control policy. Unlike previous methods that required simplified collision detection and limited torso movement, this new approach embraces stochastic contacts and extreme base rotations, enabling humanoid robots to respond dynamically to unexpected real world scenarios. And by processing discrete motion commands in real time, this framework empowers robots to perform complex full-body maneuvers that were previously unattainable, whether recovering from a fall, navigating uneven terrain, or performing natural human-like actions.

These robots are moving closer to real-world adaptability. And in another demonstration of robot dexterity, PND Botics has showcased its Adam robot performing calligraphy and other intricate physical tasks with human-like precision. The fabrication and assembly of Adam involve an extensive 317 hour process, encompassing everything from 3D modeling to final manufacturing. Boasting an impressive 37 degrees of freedom and powered by 20 actuators, Adam is designed for seamless movement and dexterous manipulation of objects in real-world industrial settings. At the heart of Adam’s performance lies the PND PMC control system, which orchestrates the operation of 163 CNC machined parts and 397 printed components.

The robot’s durability and flexibility stem from the integration of seven composite materials and 13 specialized manufacturing processes, all meticulously executed at the PND Manufacturing Center. This sophisticated engineering enables Adam to handle delicate tasks such as precision assembly, tool handling, and even tasks requiring balance and fine motor control. Despite these breakthroughs, designing a full-scale humanoid robot still presents challenges. To overcome these, PND Botics introduced a pioneering imitation learning framework, utilizing human locomotion data to refine Adam’s movements. This revolutionary approach grants Adam an unprecedented level of natural motion, allowing it to execute complex locomotion tasks with an agility never before seen in humanoid robots.

And when it comes to agentic AI breakthroughs, OpenAI just introduced deep research for chat GPT Pro users as its newest and most powerful AI tool to autonomously conduct multi-step research on the Internet. Designed to tackle complex tasks in minutes, it finds, analyzes and synthesizes hundreds of online sources into a comprehensive report. It’s powered by an optimized version of OpenAI’s upcoming O3 model, which excels in tasks like web browsing, data analysis, and real-time reasoning, while also being able to dynamically adjust its approach as it uncovers new information. As for price, chat GPT Pro’s $200 per month service looks like it may be more approachable with the inclusion of agents like Operator Orchestrator and Deep Research.

But there may be a competitor on OpenAI’s heels, as Google just unveiled its own Deep Research agent as an AI-powered tool included with Gemini Advanced, which is designed to streamline online research. Whether you’re a grad student, entrepreneur or marketer, Google’s version of Deep Research is intended to automate complex investigations by generating multi-step research plans, analyze sources and refine results in real time. Plus, it browses the web like a human, compiles key findings into a Google Doc and links sources for further exploration. As for release dates, Deep Research will first be available on mobile and in Google’s workspaces.

[tr:trw].