Insight: Figure AI's Bet on Single-Model AI
According to David Blundin, Figure AI is betting on a single-model approach for its embodied AI, a strategic choice driven by compute constraints. He argues that as embodied AI scales, local compute becomes the primary bottleneck, making a unified model more efficient than a system of multiple specialized models.
Figure's AI strategy is anchored by Helix, a unified vision-language-action (VLA) model designed to handle everything from high-level reasoning to low-frequency motor control. This contrasts with competitors like Boston Dynamics, which has traditionally used a model-predictive control (MPC) architecture for its Atlas robot, focusing on dynamic stability and motion planning before layering on more complex behaviors. Figure's approach bets that a single, end-to-end network is more scalable and will generalize faster than a system of multiple specialized models. The founder and CEO, Brett Adcock, has a track record of scaling complex hardware and software companies like Archer Aviation and Vettery. He has assembled a team with experience from Boston Dynamics, Tesla, and Apple, structuring the company to be "extremely flat" with a focus on rapid iteration—a culture where everyone either designs hardware in CAD or codes. The technical leadership has included CTO Jerry Pratt, a renowned roboticist from IHMC with deep experience in bipedal locomotion from the DARPA Robotics Challenge, though he has since left to start a new venture. In practice, Figure's single-model approach is being tested in commercial deployments. At a BMW manufacturing plant in South Carolina, its humanoids are being deployed in a phased approach, starting with identifying use cases in logistics and on the manufacturing line. The goal of this partnership is to automate tasks that are difficult, unsafe, or tedious, allowing human workers to focus on higher-value skills. This real-world application in a demanding industrial environment provides critical data for Figure's fleet learning, where insights from one robot are propagated across the entire network. While commercial manufacturing and logistics are the immediate targets, the defense sector represents a significant future opportunity. The Department of Defense has a growing interest in autonomous systems, with the Pentagon's FY2026 budget allocating $13.4 billion for AI and autonomy, and DARPA funding programs for advanced robotics. Humanoid robots are being evaluated for logistics, hazardous material handling, and reducing risk to soldiers in high-threat environments, aligning with the DoD's push to accelerate the acquisition of commercial-off-the-shelf robotic platforms.
