DARPA "Physical Intelligence" RFI: Sensing in the Structure

**DARPA** has issued a definitive Request for Information (RFI) for a new program titled **"Physical Intelligence" (PI)**. This initiative seeks to move beyond the current paradigm of "AI-as-a-brain" in a robotic body, instead focusing on materials and architectures where sensing, computation, and actuation are embedded directly into the robot's physical structure. This approach aims to solve the "latency and reliability bottleneck" that currently limits autonomous systems in high-speed, unpredictable combat and industrial environments.

The Decentralized Nervous System

Currently, a humanoid robot (like Tesla's Optimus or Figure 02) captures visual and tactile data, sends it to a central GPU/NPU for processing, and then receives motor commands. This "centralized loop" introduces milliseconds of delay—enough to cause a failure during a high-speed fall or a complex manipulation task. DARPA's PI program targets **"Smart Materials"** that can perform local, non-linear computation at the sensor level. Imagine a robotic limb where the "skin" can autonomously detect a slip and adjust the grip force in microseconds, without ever involving the central processor. This effectively mimics the human reflex system, which operates independently of the brain's conscious reasoning.

Computational Meta-Materials

The RFI specifically highlights the need for **Computational Meta-Materials**—mechanical structures that can process logic through physical deformation or electrical impedance changes. By embedding "logic gates" into the lattice of a 3D-printed robotic chassis, the robot's structure itself becomes a part of the AI inference engine. This not only reduces the weight and power consumption of centralized "brains" but also makes the robot significantly more resilient to damage. If a central processor is disabled by an electromagnetic pulse (EMP) or physical trauma, the PI-enabled limbs can still perform basic "instinctive" movements like seeking cover or maintaining balance.

The "End of the CPU" for Robotics?

While the PI program is currently in the information-gathering phase, its long-term goal is to achieve **Structural Autonomy**. This is seen as the mandatory prerequisite for the next generation of autonomous defense swarms and high-dexterity industrial robots. By moving the compute to the point of contact, DARPA aims to create machines that move with the fluid, reactive grace of biological organisms. The deadline for RFI responses is May 27, 2026, with the first prototype contracts expected to be awarded by early 2027.

As **Physical AI** transitions from software world-models (like NVIDIA Cosmos) to structural reality, the DARPA initiative proves that the ultimate machine won't just have a brain—it will *be* a brain. The separation between software and hardware is finally dissolving.