Figure founder and CEO Brett Adcock has unveiled Helix , a new machine learning model for humanoid robots . It is a “generalist” Vision-Language-Action (VLA) model.
VLAs are a new trend in robotics, leveraging visual and language commands to process information. Currently, the best-known example is Google DeepMind's RT-2, which trains robots using a combination of video and large language models (LLM). Figure's Helix works in a similar way, combining visual data and language prompts to control a robot in real time.
In an ideal world, we would be able to tell a robot to do something and it would just do it. According to Figure, that’s where Helix can help. The platform is designed to bridge the gap between vision and language processing. After receiving a natural language voice command, the robot would visually assess its environment and then perform the task.
See also: Meet Adam: An impressive barista robot!
Figure gave two example commands, “Give the bag of cookies to the robot on your right” and “Take the bag of cookies from the robot on your left and place it in the open drawer.” Both of these examples involve a pair of robots working together. Helix is designed to control two robots simultaneously, with one helping the other perform various household tasks.
Figure introduces the VLM model, highlighting the work the company has done with its 02 humanoid robot in the home environment. Homes are more challenging for robots, as they lack the structure and consistency of warehouses and factories. Difficulty with learning and control are significant obstacles. These issues, along with high prices, are why home robots are not a priority for most companies that make humanoid robots.
See also: Humanoid robots dance with human movements with ExBody2
With the announcement of Helix, Figure is making it clear that housework should become a priority for humanoid robots. However, there are many challenges. Training robots to do complex tasks in the kitchen, for example, requires a wide range of actions with different settings.
“For robots to be useful in households, they need to be able to create intelligent new behaviors on demand, especially for objects they’ve never seen before,” says Figure. “Teaching robots even a new behavior currently requires substantial human effort: hours of programming or thousands of demonstrations.”
Manually scheduling tasks at home is difficult because there are so many unknowns. Kitchens, living rooms, and bathrooms are all dramatically different. So are the tools used for cooking and cleaning. Plus, people leave clutter, rearrange furniture, change lighting. All of this takes hours of training.
See also: Pioneering half marathon in China: Humans vs Robots
Figure's release of Helix is a promising development in the field of humanoid robots. With this new machine learning, robots can now understand and perform complex tasks with ease.
One of the key challenges in human-robot interaction has been the lack of natural communication. Traditional programming methods require specific commands, limiting its flexibility and adaptability in dynamic environments.
Source: techcrunch.com
