The development of humanoid robots capable of training and fighting on their own, without direct human control, is advancing quickly. In 2026, a major breakthrough came with the introduction of "foundation models" designed for physics. These robots can now anticipate, dodge, and strike in real-time, without needing a human operator. While initially developed for combat applications, these robots' abilities—such as fast reflexes and predictive movement—can also be used in various other fields. One example is the work by Unitree, a robotics company that demonstrated a robot capable of thinking ahead during a fight. Previously, their boxing robot needed a human operator using a joystick or control system. Now, the new UnifoLM-X2-1.0 system allows the robot to control its actions independently. Instead of just reacting to an opponent's punch, the robot builds a mental model of the situation, allowing it to make more strategic moves. A physics-informed foundation model, also known as a World Foundation Model, is a type of artificial intelligence trained on the fundamental laws of physics—such as gravity, friction, and motion—and on extensive simulations or real-world demonstrations. These models act as both the brain and motor system for humanoid robots. Unlike traditional AI, they give robots a deeper understanding of how the physical world works, enabling them to learn complex tasks and react smoothly in real-time, such as dodging or striking without human input. Traditionally, each movement or action a robot performed required its own specific algorithm. However, physics foundation models are changing this by integrating physical equations directly into their learning process. This means the robot can understand and predict how forces like gravity and momentum affect its movements. For instance, if the robot is hit or slips, it can instantly adjust its balance to prevent falling. These models are also trained in highly realistic virtual environments, which allows robots to transfer their skills to the real world without needing additional training. This makes the development process faster and safer. These advanced robots rely on high-performance motors and actuators to move and react quickly. For combat, they need powerful joint torque, often with more than 17 to 30 degrees of freedom. These motors, developed by companies like Faulhaber, allow for ultra-fast control. Inside the robot’s body, powerful CPUs, AI accelerators, and GPUs handle real-time processing of visual data and movement commands. The robots also use sensors such as LiDAR and tactile skins to detect impacts and understand their surroundings. Several companies and research institutions are leading the way in this field. Unitree Robotics, for example, has developed open-source foundation models that can control humanoids in real-time. Another company, Physical Intelligence, has introduced π0, a model that gives robots the ability to manipulate objects and move smoothly. NVIDIA has also launched GR00T, a humanoid foundation model that works with SONIC, a system that controls entire body movements for dynamic actions like running or kicking. In September 2026, Unitree made headlines by unveiling the G1, a humanoid robot capable of sparring with a human in real-time without any human input. This marked a major shift from earlier demonstrations that relied on remote controls or VR headsets. Following this, the official newspaper of the Chinese military called for the rapid adoption of these technologies for military training. Meanwhile, academic research continues to push the boundaries, with projects like RoboStriker and BeyondMimic enabling robots to learn complex movements autonomously. These robots have the potential to transform industries such as defense, disaster response, and industrial maintenance. They could replace humans in dangerous tasks like urban warfare, mine detection, or surveillance in war zones. They are also designed to be programmable through APIs, similar to robot dogs used in military operations. As the technology continues to evolve, the integration of these robots into various sectors is expected to grow significantly.