① MIT's Neural Jacobian Fields (NJF) enables robot motion control through visual self-observation, eliminating reliance on physical sensors or digital models;

② The system maps motor signals to movements using multi-camera training data, achieving real-time 12Hz control with a single camera post-training;

③ Demonstrated on soft grippers and robotic hands, it shows adaptability but requires per-device training and currently lacks tactile feedback capabilities.

➀ Researchers developed a TAMP-based algorithm (cuTAMP) that uses GPU-accelerated parallel computing to evaluate thousands of packing solutions simultaneously;

➁ This method reduces decision-making time to seconds, enabling efficient packing of varied objects in tight spaces without prior training;

➂ Demonstrated in real robots and simulations, the approach shows versatility for industrial tasks and potential integration with AI language models.

➀ Researchers developed TWIST, a real-time teleoperation system enabling humanoid robots to mimic full-body human movements for tasks in inaccessible or dangerous environments;

➁ Tested on Unitree's G1 and Booster Robotics' T1, TWIST achieves human-like coordination by leveraging motion data from legs, feet, and other body parts;

➂ Future enhancements aim to reduce reliance on motion-capture systems and enable autonomous task learning for broader industrial and hazardous applications.

➀ Researchers at the University of Coimbra developed a novel soft actuator powered by boiling water, converting water into steam via an embedded heater to generate motion, offering simplicity and scalability without pumps or high voltage.

➁ The actuator operates at 24V, delivers over 50N of force, and achieves rapid pressure build-up (100 kPa/s), demonstrating durability through 1,000+ cycles and outperforming traditional designs in speed and reliability.

➂ Prototype robots, including a gripper and a climbing robot (Bixo), showcase real-world applicability for tasks like object manipulation and navigating unstructured environments, highlighting safe, low-cost, and pump-free solutions.

➀ MIT developed a "Relevance" system to help robots prioritize tasks using audio-visual inputs and predict human intentions for efficient assistance;

➁ The system achieved 90% goal prediction accuracy and 96% correct object selection in breakfast scenario testing, with 60% fewer safety incidents;

➂ Inspired by human brain's RAS mechanism, the robot dynamically switches between observation mode and active assistance mode based on environmental context recognition.

➀ Researchers at UC San Diego's Bioinspired Robotics Lab have developed 3D-printed robots capable of walking without electronics, powered by compressed gas.

➁ These robots are designed to operate in extreme environments like high radiation zones, disaster areas, and space missions, showcasing continuous operation for three days when connected to a gas supply.

➂ The robots feature a monolithic design printed in one piece using flexible materials, with future plans to integrate gas storage and biodegradable components for enhanced functionality.

➀ Researchers from Carnegie Mellon University and NVIDIA have developed a new model to train robots to move like human athletes.

➁ The model, called ASAP, involves a two-stage training framework that improves the adaptability and fluidity of robot movements.

➂ While the robot successfully replicated several iconic sports moves, more work is needed to make robots as athletic as professional athletes.