Summary
Transcript
One X technologies just released another incredible demo of their Eve humanoid robot, learning to do a series of tasks by just watching, but not how youd probably expect. See, with the culmination of several recent breakthroughs, artificial intelligence can both understand and generate human like language. But by finally giving these state of the art AI systems a mechanical, human like form, the Eve Android can put them all into practical use for the first time.
And it all starts by using NeOs neural vision system to translate all observed actions into controlled motor movements, allowing the robot to complete everyday tasks like tidying a room, making a cup of coffee, doing laundry, and more. But to seamlessly work in these environments. The Eve humanoid is strategically designed to stand 186 cm tall, so it can most easily maneuver through doorways, hallways, and rooms while completing chores and assisting in the same areas humans occupy.
Plus, the Androids 86 kilogram frame uses muscle like mechanical actuators to safely interact with people and lift objects that weigh up to 20. With its six hour battery life, with a walking speed of 14. 4, Eve robot could potentially walk a distance of several kilometers before needing to return to its charging dock. And while it’s made for assisting humans, the robot will still face situations where it needs a helping hand, like getting stuck or being unable to complete a certain task.
In these cases, a remote human operator can teleoperate the Android. Otherwise, one X can just upload its updated skill set to expand its capabilities. To do this, the robot uses its pioneering reinforcement learning system developed by one X, which is a massive dataset of 30 different robots performing a huge range of physical tasks like organizing, picking up objects, and socially interacting. First, this data is used to train the robots base AI model to understand general physical behaviors and skills.
Next, the AI model is fine tuned into more specific capability groups like door manipulation or warehouse tasks. Then it goes through a final tuning process to align its skills for performing any specialized tasks in real world environments. Altogether, it serves as a streamlined training system, which allows anyone to rapidly expand on what the neo robot can do in just minutes, all while using a desktop gpu instead of needing the demanding development time and compute resources of the past.
And one X says that all of the real time learning capabilities it demonstrated were all achieved through this new training approach, with no pre programmed scripts or computer graphics required. This allows for the utmost fluidity of movement as the robot’s vision based control system refreshes ten times per second to control all movements. And while no release date has been set for consumer availability, the Eve robot provides a fascinating glimpse into an AI driven future where machines learn like humans instead of like computers.
Meanwhile, another humanoid breakthrough for healthcare was just showcased by a chinese tech company with another new robot that’s solving the world’s oncoming caregiver crisis. Meet GR one, an advanced humanoid robot from Fourier intelligence that is impressing crowds with its autonomous abilities. Standing 165 cm tall and weighing 55, GR one moves with remarkably human like motion, being able to walk, avoid obstacles, or even lift and transport elderly patients with ease.
But even more impressive is the robots ability to be further programmed for all kinds of tasks, like sitting, standing, jumping, and using tools or utensils. To this point, the company says its GR one is bringing about a new era of robotic caregiving technology with several versatile capabilities. For example, the robot’s artificial intelligence allows it to communicate to the elderly just like a person. In fact, the GR one can have natural language conversations, provide useful information, offer comfort, and even respond with empathy.
This is a future where the androids could gently wake up a grandma, explain her therapy routine, and keep her company over breakfast with friendly discussion. This blend of physical care with emotional warmth could vastly improve quality of life for millions. And while the GR one’s current skeletal form looks a bit creepy with exposed wires, Fourier plans to give the final version a sleek chrome casing for a much friendlier vibe in healthcare settings.
But that’s not the only demonstration of an AI robot assistant using end to end learning, as Menti Robotics has also just unveiled their latest humanoid demonstration with a twist. After two long years of stealth development, the highly anticipated menti bot was unveiled as an advanced humanoid robot that takes a radically new approach towards intelligent Android assistance. And while the company has raised $17 million in funding, they’re still in the prototyping stage of designing the various AI systems that allow it to dynamically understand commands, devise a plan of action, and integrate seamless locomotion and dexterity to complete a wide variety of household tasks and chores.
In fact, the researchers believe that by combining computer vision, natural language processing, highly realistic simulation, and techniques for transferring between virtual and real world environments, it could serve as a launching point for creating future general purpose bipedal robots that can navigate freely like humans do, while having the intelligence to tackle household tasks via imitation learning. Furthermore, the robot uses large language models based on the transformer architecture to interpret spoken or written instructions and dynamically map out the steps required to execute commanded tasks.
Plus, its AI reasoning is paired with advanced computer vision to analyze its surroundings and state of the art robotics to dexterously manipulate objects while maintaining dynamic balance when carrying loads and menti is aiming big first by targeting both industrial and home settings right out of the gate, which is an ambitious goal as most humanoid robot makers focus first on structured industrial environments before rolling into the household. And finally, in perhaps the most intriguing development, a multinational collaboration of 13 leading institutions may have cracked the code on training robots for human like intelligence and dexterity in the real world.
Their secret an innovative new dataset called Droid that exposes robotic systems to over 1400 unique camera viewpoints capturing intricate robotic arm movements and object interactions. In fact, droid is short for dexterous robotic interaction operating dataset, and it has the potential to vastly enhance robot performance across manipulation tasks in diverse environments. Its key advantage is its use of a unified hardware setup for data collection that spans lab, office, and even household settings.
As for its makeup, the highly portable droid apparatus consists of a seven degree of freedom robotic arm mounted to an adjustable desktop alongside stereo cameras and VR teleoperation equipment. This flexibility allows the researchers to rapidly modify data collection scenarios while maintaining standardization, and early results suggest exposing robots to droid yields significant functional gains. In fact, the policies learned using this diverse dataset demonstrated up to 22% higher task success rates on average compared to other state of the art approaches.
Plus, droid trained robots also exhibited smoother motion planning and improved spatial reasoning when faced with changing conditions or novel objects. In conclusion, the innovative droide dataset promises groundbreaking advancements in training robots for human like intelligence and dexterity. With its diverse collection of real world scenarios and standardized hardware setup, Droid has shown remarkable potential in enhancing robot performance across various environments. .
