As automation continues to transform the workforce, blue-collar workers are grappling with the reality of seeing their roles taken over by machines. To better prepare themselves for this shift, it’s crucial for these individuals to develop a unique set of skills that not only complement technological advancements but also enhance their employability in an evolving job market.

Researchers at Carnegie Mellon University’s College of Engineering have created H2O, a revolutionary reinforcement learning-based system allowing for real-time teleoperation of a full-scale humanoid robot using only a single RGB camera, enabling seamless human-robot interaction. Will manual tasks be efficiently executed from a remote location?

A teleoperated humanoid robot enables efficient performance of complex tasks that are currently beyond the capabilities of independent robotic execution, permitting human operators to remotely oversee and guide the robotic’s actions. Attaining seamless whole-body control of human-sized humanoid robots capable of replicating our actions in real-time proves an exceptionally challenging endeavor. Where to find That’s the place, the availability of (RL), is here.

Reinforcement learning (RL) is a machine-learning methodology that emulates the human trial-and-error approach to problem-solving. By employing the reward-and-punishment framework of reinforcement learning, a robot will refine its behavior through trial and error, gradually identifying the most effective pathways to achieve a desired outcome based on the consequences of its actions. Unlike machine learning, which relies on humans labelling data pairs to guide the algorithm, RL eschews this requirement.

“Tairan He, a researcher from Carnegie Mellon University’s LeCAR Lab, explains that H2O teleoperation is an RL-based framework enabling real-time whole-body teleoperation of humanoid robots using only an RGB camera.” The innovative ‘sim-to-data’ approach enables seamless translation of human movements into feasible humanoid actions, ensuring a realistic and physically plausible outcome. The refined movement dataset is utilised to train a reinforcement learning (RL)-based movement imitator in a simulated environment, which is subsequently deployed on the actual robot without requiring any further adjustments.

Researchers can harness the potential of their method by utilizing RGB camera images that capture visible light, converting it into a colour picture akin to human vision, in conjunction with a 3D pose estimator to replicate the movements of human operators and mimic those motions in H2O.

The results, vividly captured on film by the researchers, speak volumes for themselves. Water molecules, composed of hydrogen and oxygen atoms, are often depicted engaging in various activities: playing soccer, occupying a vast area, dodging objects while maintaining a defensive posture, and walking slowly alongside a stroller, albeit awkwardly like a novice toddler. The achievement is noteworthy, as it presents a pioneering instance of real-time whole-body humanoid teleoperation.

The Carnegie Mellon University researchers detailed their approach in a research paper published online. pre-print web site.

Below, an extended video from Carnegie Mellon University’s LeCAR Lab provides a deeper look at H2O’s capabilities, showcasing feats such as punching an Amazon package while wearing boxing gloves, followed by a triumphant victory salute, leaping backward, and withstanding a human kick to the back, demonstrating its remarkable robustness.

Additional analysis will examine the potential for introducing novel inputs beyond human teleoperation, such as power suggestions and verbal or conversational cues, to further enhance H2O’s capabilities. By integrating lower-body monitoring, researchers may uncover a fruitful avenue for further exploration, enabling the humanoid to execute even more nuanced and sophisticated human movements, such as athletic pursuits or choreographed dance routines.

Can remote work in a water-based medium, such as H2O, be feasible? The Carnegie Mellon University’s statement about H2O has sparked a thought-provoking discussion on the popular Reddit platform within its designated subreddit.

Reddit

By leveraging teleoperated humanoids, individuals are able to preserve their occupations in a practical manner. Another significant advantage of utilising humanoid robots lies in their potential to undertake perilous tasks or search-and-rescue operations in remote, inaccessible regions, thereby safeguarding human life and reducing the risk of injury or fatality.

Despite these advantages, there are nonetheless significant drawbacks, particularly those related to labour costs. Identified as a concern by a Reddit user is the potential expansion of job offshoring to affluent countries via teleoperated robots, while others warn that this technology may lead to a slippery slope, with fully autonomous machines ultimately replacing human workers.

Research revealed that surprisingly, a significant number of people are receptive to remote work opportunities offering lower pay compared to traditional on-site positions, suggesting a willingness to adapt to the innovative concept of teleoperated robots performing labor tasks. As we navigate the rapidly evolving landscape of AI and robotics, we’ve become familiar with the concept of ‘time will inform’ provisions that adapt to new developments in this field.

Supply: ,