Modern robots have made impressive strides, mastering a variety of tasks such as mixing cocktails, running marathons, and folding laundry. However, they still fall short in executing a range of actions on command. This gap was evident at the recent Robotics Summit held in Boston, where promotional materials clashed with the reality presented by robot creators, as reported by Tech Xplore. Elon Musk showcased his Tesla Optimus prototype, which recently learned to take short steps, while Chinese companies like AgiBot and Matrix Robotics demonstrated robots capable of greeting guests, serving coffee, and giving tours. Despite these advancements, the capabilities of contemporary robots remain limited. Chris Matthieu from RealSense noted that most humanoid robots are either remotely operated or designed for very specific tasks and paths. The 1X Neo, touted as the world's first consumer-ready humanoid robot aimed at transforming home life, was controlled by a person during its demonstration. Nevertheless, progress is being driven by advances in artificial intelligence. William Okazaki from Renesas highlighted that AI has significantly accelerated growth in this field. The long-standing challenge of robotic dexterity is being addressed, as robots can now grasp objects with high precision, and sensors can even detect when they make contact with human skin. This progress has largely been supported by AI models focused on "vision-language-action" (VLA), which combine real-time camera images with written instructions, allowing robots to link what they see with what they need to do. Additionally, "world models" enable AI to learn from extensive images and videos, predicting future behaviors of objects based on their interactions. However, it will still take years before a humanoid robot capable of performing a wide range of tasks autonomously becomes a reality. Daniel Fan from Innodisk stated that more time will be needed for general-purpose robots to develop. While humanoid robots like Boston Dynamics' Atlas and Hexagon Robotics' AEON are operational in real-world settings, they are still in the testing phase rather than commercially available. Charlie Kemp from Hello Robot cautioned that understanding a robot's true capabilities requires observing how it attempts tasks. Full-scale autonomous operation remains elusive due to insufficient data, as Xinrui Bi from AgiBot pointed out. To gather more data, companies are deploying cameras to monitor human movements in various scenarios, from home cooking to textile workshops in India. The stakes are high, as errors in physical environments can lead to real harm, emphasized Valentino Fagard from XELA Robotics, who is working to enhance robots' tactile feedback. While engineers can impose limitations on robots' actions, predicting their behavior remains challenging due to their inherent unpredictability. John Black from Brain Corp explained that the nondeterministic nature of these systems makes it difficult to anticipate outcomes. As robots continue to evolve, the market will see increasing competition, with companies striving to enhance safety and functionality in the realm of robotics.
Informational material. 18+.
