Figure 02 Humanoid Robot Demo: Package Sorting & Future of AI with Brett Adcock

Figure 02 Package Sorting Demonstration

• 💡 The video showcases a Figure 02 humanoid robot performing small package logistics for a full 60 minutes, uncut, demonstrating its ability to identify, pick, and orient packages for scanning.
• 🎯 The primary task involves handling everyday packages, including challenging deforming plastic bags and boxes, requiring the robot to flatten bags or find barcodes on six sides.
• ✅ The goal is to achieve a high 90s% success rate at a speed of 3-4 seconds per package, with the robot continuously running since 9:30 AM on the day of recording.

Advanced AI & Performance

• 🧠 Figure's approach relies on a single neural network, Helix S1, running onboard the robot, which reasons over camera frames to command full actions.
• 🚀 Key model improvements include stereo camera images, force feedback, temporal memory (several seconds of camera frames), and a larger parameter size model, boosting both speed and success rate.
• 📈 The robot's speed has significantly increased, with the demo showing 4.05 seconds per package, and recent internal tests achieving 3.54 seconds, with a target of under 2 seconds by year-end, surpassing human performance.

Humanoid vs. Specialized Robotics

• 💡 The task of sorting packages is currently performed by humans, who have a roughly 90% success rate but experience high turnover and require breaks, unlike robots that can work 24/7.
• 🤖 Figure emphasizes the development of general-purpose humanoid robots that can adapt to various tasks (logistics, manufacturing, construction, home) rather than specialized, single-use automation.
• 🎯 The humanoid form factor is considered ideal because the world is designed for humans, allowing for efficient transfer learning from human demonstrations and serving as the ultimate deployment vector for Artificial General Intelligence (AGI).

Figure 03 & Future Scaling

• 🛠️ The upcoming Figure 03 robot features significant physical enhancements, including doubled finger speeds, custom-designed cameras, and improved hands, aiming for human-like range of motion, speeds, and payloads.
• 🚀 Figure 03 has already achieved its first fully autonomous use case in the lab, demonstrating a much faster bring-up process compared to previous versions, being several months ahead of schedule.
• 🏭 The company is focused on scaling production to thousands of robots at their Bot Q facility, with Figure 03 designed for high-rate manufacturing and no current supply chain issues.

Vision for Autonomous Work

• 🌟 Figure's core mission is to "sell humanlike work" through fully autonomous agents that provide clear ROI for customers, avoiding non-essential features like "box jumps or backflips."
• 📊 Robot mistakes, such as being "flummoxed," are attributed to being "out of distribution" and are resolved by feeding more data into the neural network, creating a data flywheel where deployed robots gather data for continuous improvement.
• 🔑 The future of robotics hinges on co-designing hardware and AI to create smarter, cheaper robots at scale, with Figure aiming for a first-mover advantage by rapidly deploying millions of units.