\boxed{“`html

How Figure AI’s “Learned Natural Walking” Advances Humanoid Robot Motion

Humanoid robots have made incredible strides in recent years, demonstrating an array of athletic feats that many humans could only dream of. Whether it’s running, jumping, breakdancing, or punching, these bipedal machines can perform a wide range of complex movements. However, one area where humanoid robots have traditionally struggled is in something seemingly mundane yet crucial: walking a straight line naturally. For many robots, the act of walking often appears stiff and awkward, almost as if they are “moments away from soiling themselves.” Enter Figure AI, a leading startup in the humanoid robot space, which claims to have engineered a groundbreaking solution to make their robots’ gaits more natural.

Figure AI has introduced what they refer to as “Learned Natural Walking”, a new technique that makes their Figure 2.0 humanoid robot walk in a way that more closely resembles human movement. To achieve this, the company analyzed thousands of virtual humanoid robots simultaneously walking in a simulated digital environment. Through the use of reinforcement learning, they rewarded virtual robots for actions such as synchronized arm swings, heel strikes, and toe-offs (when the toe leaves the ground). According to Figure AI, this simulated training environment allows them to “generate years worth of data in just days.” The resulting “Learned Natural Walking” model was then applied to a physical Figure 2.0 humanoid.

Working Towards Real-World Walking

While the new stride isn’t flawless, it marks a significant improvement compared to what the company demonstrated merely a month earlier. Figure AI recently shared a video showcasing their robots’ new “learned natural walking” model. The latest version shows a more natural movement, particularly in the heel and lower leg actions. By exposing their army of simulated robots to various terrains and unexpected scenarios like tripping or being shoved, Figure AI believes that this training makes strides towards making their robots adapt better to real-world, unpredictable environments.

“These initial results are exciting, but we believe they only hint at the full potential of our technology,” Figure AI stated in a blog post. “We’re committed to extending our learned policy to handle every human-like scenario the robot might face in the real world.”

Research indicates that human-like appearances and conversational abilities make people more receptive to robots. Such human-likeness also makes it less likely that people would want to harm the machines—an important consideration given the increasing push for deploying humanoid robots across various industries. Figure AI has already signed a deal with BMW to test its humanoid robots in a manufacturing facility in South Carolina. Similarly, Amazon is reportedly trialing humanoid robots from another robotics company, Agility Robotics, in some of its warehouses.

Robots Are Finally Getting Better at the Little Things

Humanoid robots have been performing backflips and executing dance routines for years now, yet they often stumble when it comes to simple human tasks. This phenomenon—where robots find challenging what humans find easy and vice versa—is known as “Moravec’s Paradox”. Although advances in AI have helped narrow this gap, there is still a considerable way to go.

For instance, Google DeepMind made a significant breakthrough last year by teaching a robot how to tie a shoe. However, even after decades of robotics research, the robot performed the task less proficiently than a school-age child. Moreover, robots today still struggle with tasks such as lightly gripping objects without crushing them or maintaining their balance as effectively as humans do.

Conclusion

In this context, Figure AI’s “Learned Natural Walking” model represents a significant step forward in making humanoid robots more adept at everyday movements. By leveraging advanced reinforcement learning in a simulated virtual environment, Figure AI has made strides in making their robots walk more like humans. Although the results are still a work in progress, the company’s efforts mark a promising path toward making humanoid robots more functional and relatable in real-world settings.

With partnerships such as the one with BMW already in place, Figure AI is well on its way to making their humanoid robots a common sight in various industries. While challenges still remain—such as mastering tasks that humans find simple—advances like “Learned Natural Walking” bring us one step closer to a future where robots can seamlessly integrate into human environments.

“`}



Source link