- Published on
Figure's 'Most Boring Video' is a 60-Minute Statement on Humanoid Endurance
- Authors
- Name
- Humanoids daily
- @humanoidsdaily
Figure AI's 'Most Boring Video' Aims to Prove a Critical Point: Endurance
Sunnyvale, CA – In a strategic move away from short, curated clips, Figure AI has released a full, 60-minute video of its Figure 02 humanoid performing a complex logistics task without interruption. CEO Brett Adcock introduced the footage by calling it "the most boring video we've ever posted," a deliberate framing to shift the narrative from flashy demos to the kind of sustained, useful work required for real-world commercial deployment.
"I want to emphasize: Figure is here to deliver humanoid robots - at scale - that do real, useful human-like work," Adcock stated in a thread on X. "That means running for hours every day and consistently hitting speed and performance targets. That’s what matters to our customers."
The video shows the robot handling a continuous stream of varied packages, a task Adcock insists is impossible to solve with traditional, hard-coded robotics due to the immense variability. The demonstration serves as a testament to the progress of Figure's end-to-end neural network, Helix.
Welcome to the most boring video we've ever posted Here's 60 minutes of our humanoid robot solving logistics, powered by our Helix neural network
Helix Gets Smarter: Memory, Touch, and Data
Alongside the video, Figure published a detailed technical blog post outlining the significant upgrades to the Helix AI model that enabled this performance leap. The improvements stem from both scaling up training data and enhancing the model's core architecture.
According to Figure, the key architectural improvements include:
- Temporal Memory: Helix now has a short-term visual memory, allowing it to compose information from a sequence of recent video frames. This gives the robot a sense of context, helping it remember which side of a package it has already inspected, eliminating redundant movements and improving strategic manipulation.
- Force Feedback: The model now integrates force sensing into its state input, giving the robot a proxy for a sense of touch. This allows Helix to "feel" when it makes contact with an object or surface, leading to more precise grips and better control.
- State History: The AI's input now includes a history of the robot's own recent states (hand, torso, and head positions), enabling faster, more reactive control and better continuity between actions.
These model improvements were combined with a significant increase in training data. Figure revealed that scaling training from 10 to 60 hours of human demonstrations directly resulted in a drop in average package handling time from 6.8 seconds to 4.3 seconds, while barcode scanning accuracy climbed from 88% to over 94%.
Measurable Leaps in Performance
The result of these combined efforts is a more capable and efficient system. Compared to just a few months ago, Figure's logistics model now demonstrates:
- Handling of New Package Types: The robot now reliably manipulates challenging deformable poly bags and flat envelopes.
- 20% Faster Throughput: Despite the increased complexity, the average handling time has improved to approximately 4.05 seconds per package.
- Higher Scanning Success: The barcode is now correctly oriented for scanning around 95% of the time.
- Adaptive Behaviors: The robot has learned subtle, human-like techniques, such as patting down wrinkled plastic mailers to ensure a good barcode scan, a behavior learned directly from demonstration data, not explicitly programmed.
The company also noted the model's ability to adapt and re-plan after a mistake, such as a failed grab, highlighting the robustness of the learning-based approach.
A Focus on Real-World Deployment
This long-form demonstration is a clear statement of intent from Figure. By focusing on endurance, reliability, and measurable performance metrics, the company is directly addressing the core requirements of its commercial partners in logistics and manufacturing.
Adcock ended his announcement with a forward-looking statement and a call for talent. "2025 will be a big year! We’ve kicked off production, we’re shipping more robots to customers, and we’re work[ing] hard on robots in the home," he wrote, reinforcing the company's aggressive timeline and expanding ambitions.