In a significant development for the robotics industry, Figure, a leading Bay Area company, has unveiled a new machine learning model designed specifically for humanoid robots. Announced by Brett Adcock, the CEO of Figure, the model is called Helix, a “generalist” Vision-Language-Action (VLA) system. This announcement comes just two weeks after Figure declared its intention to develop in-house models instead of collaborating with OpenAI.

Vision-Language-Action (VLA) models represent a cutting-edge approach in robotics, utilizing both vision and language commands to process and act on information. The most well-known example of this technology is Google DeepMind’s RT-2, which leverages video and large language models (LLMs) to train robots. Helix functions similarly, as it integrates visual data and language prompts to control a robot in real time.

How Helix Works

Helix’s ability to generalize objects is one of its standout features. According to Figure, “Helix displays strong object generalization, being able to pick up thousands of novel household items with varying shapes, sizes, colors, and material properties never encountered before in training, simply by asking in natural language.” This means you can tell a robot to perform a task, and it will execute it accurately based on visual and language prompts. Helix not only simplifies the user’s task but also ensures the robot can handle and process objects with complex dimensional makeup.

Applying Helix in Everyday Settings

Helix’s real-world application is being showcased with Figure’s 02 humanoid robot in home environments. This represents a monumental step in domestic robotics, considering the diverse and unpredictable nature of American households. Compared to structured warehouses, homes pose a greater challenge with their irregular layouts and varied objects. However, Helix’s adaptability makes it well-suited for these settings.

“Now, for robots to fully get going in our daily lives, they require to create intelligent actions in-context for a range of tasks that will come their way.”

Figure

Weather be it designing a perfectionate meal or trivial routine tasks like turning off the lights – Implementation of robots in our home life is a few years away because currently learning to recognise these variables.

Economics and the Future of Robotics

Despite a great deal of promise, we are still far from robotics redefining our home lives economically. Families prefer a range of different from that of industries and that too with performing minimal unrealistic intellectually demanding tasks. Before significantly stepping up our expectations, we need a better understanding of this field along with solving these multifaceted challenges. Besides, the learning process for robots in dynamic domestic settings requires large hours of drills and monetary upscale factors from infrastructure to manpower.
Hence, all major companies focus first on building the reliability and reducing the cost in less dynamic stable Industry settings.

Hurdles in Program vs. Training

Affforming a robot to do an unfamiliar task is a major challenge for addressing adaptability. While Manual programming saves time and resources but won’t scale as per America’s dynamic lifestyle. To affirm Helix realtime capabilities, it was tested in an elaborate manufactureable environment.

Expanding Insights and Looking Forward

Figure has spent considerable resources in enlarging its team to make breakthrough leaps. However, it is a partial glimpse of segment on advanced robotics environment.