Figure has transitioned from manually coding robot behaviors to using a proprietary vision-language-action neural network called Helix. This model runs inference directly on the robot's on-board GPUs, enabling autonomous operation without continuous connectivity.
Figure is developing new robot generations approximately annually, achieving a 90% cost reduction between its Figure 2 and Figure 3 models. The current production cost is well under $100,000 per unit, with the upcoming Figure 4 expected to be another major leap.
Figure is proving its technology in demanding industrial environments, such as its six-month deployment on a BMW factory line. The company plans to leverage these commercial successes to eventually enter the consumer market with a Robotics-as-a-Service (RaaS) model, leasing robots for $400-$600 per month.
With the hardware becoming increasingly robust, Figure identifies the lack of diverse, high-quality training data as the main obstacle to achieving widespread deployment and generalization. The company is heavily invested in large-scale data collection efforts to train its Helix model.
Figure's leadership believes that artificial general intelligence (AGI) may be achieved first in embodied forms like humanoids, rather than purely in digital models. This philosophy drives their focus on creating a general-purpose physical body for advanced AI.
Keep pulling the thread on White House.