Figure's Brett Adcock: Deleting Code for Neural Net Humanoids & Autonomous Manufacturing

Paradigm Shift in Robotics Control

• 💡 Figure's groundbreaking approach involves eliminating 109,000 lines of C++ code, replacing explicit programming with a neural-net-based "System Zero" controller.
• 🧠 This shift is driven by the mathematical impossibility of programming all potential states for a complex humanoid robot, emphasizing that intelligence emerges from data volume, not programming complexity.
• 🚀 The new paradigm moves from traditional software to intuition born from data, fundamentally inverting foundational assumptions in robotics.

Skill Acquisition & Fleet Learning

• 🎯 Robots learn tasks through human tele-operation demonstrations, where human guidance generates the data needed for the neural net to derive autonomous policies.
• ✅ A key disruption is that once a single robot in the fleet learns a task, the entire fleet knows it instantaneously, transforming labor into a permanently compounding asset.
• 🌱 This method bypasses complex physics modeling, focusing instead on the grunt work of gathering sufficient human demonstration data.

Challenging AI & Hardware Assumptions

• ⚠️ Brett Adcock delivers a sharp reality check, stating that Large Language Models (LLMs) lack grounded understanding of physics, often