Rethinking Robotics with Physical Intelligence

Rethinking Robotics with Physical Intelligence is sponsored by DARPA - Defense Advanced Research Projects Agency. Seeks to embed intelligence directly into the physical materials of robotic systems, enabling real-time response without relying on continuous external computation or communication links.

Rethinking robotics with physical intelligence | DARPA Department of War organization. Rethinking Robotics With Physical Intelligence Rethinking robotics with physical intelligence RFI and workshop to explore embedded compute and multi-functional materials (Thermoplastic) Flower in the Sun. Stimuli-responsive polymers have gained increasing attention for their applications ranging from soft robotic grippers to actuators.

By controlling strain within thin thermoplastic sheets, these small grippers can transform into three-dimensional shapes based on a photothermal response and withstand loads more than 24,000 times their own mass. DARPA’s new RFI seeks a new class of robotics materials capable of intermixed sensing, adapting, and acting in real time without relying on continuous external computation or communications links.

Source: Air Force Research Laboratory | Courtesy photo Amber Hubbard Today’s advances in robotics are often driven by breakthroughs in artificial intelligence, machine learning, and perception. But in complex and constrained environments, the limiting factor is often hardware, not software.

Systems that rely on constant data processing, high-bandwidth communication, and centralized compute can face delays, power constraints, and vulnerabilities that limit performance or prevent mission success altogether. DARPA is looking to tackle these challenges by embedding intelligence directly into the physical materials of robotic systems.

A new Request for Information (RFI), calls on the research community to help define a new class of materials capable of intermixed sensing, adapting, and acting in real time without relying on continuous external computation or communication links. While the RFI itself is exploratory, it is a first step toward a more immediate opportunity: an invite-only, in-person workshop planned for the summer 2026.

Selected participants will have the chance to present their ideas, engage with DARPA, and inform future program directions. Rethinking where intelligence lives Commercial robotics has largely centered on building systems that can operate alongside people, often emphasizing familiar shapes and interfaces. National security applications demand something different.

Robotic systems for defense must operate in extreme, unpredictable, and adversarial environments with limited communication and little opportunity for human intervention. In these conditions, performance is not defined by how much data a system can process, but by how quickly and reliably it can respond. Meeting these demands requires a shift in where intelligence resides.

DARPA is exploring physical intelligence, an approach that embeds sensing, computation, and actuation directly into materials, components, and structures. Instead of routing information through centralized processors, future systems could respond through their physical design, enabling faster, more efficient, and more resilient operation in dynamic environments.

“Today’s robots are often limited by the need to sense, process, and act as separate steps,” said DARPA Program Manager Julian McMorrow. “We are interested in collapsing that loop by embedding intelligence directly into the hardware, so systems can respond in real time without relying on constant data movement.

” This shift could enable robotic systems that are faster, more energy efficient, and more resilient in complex, unstructured environments.

A focus on materials, not machines The RFI targets foundational advances at the material, component, and kernel level, with an emphasis on two areas: Actuation and sensing: DARPA is interested in materials and structures that integrate sensing, actuation, and even elements of control into the same physical substrate, enabling robots to perceive and interact with their environment with greater speed, efficiency, and adaptability.

Dynamic and adaptive closed-loop compute: Rather than relying on centralized processors and large data flows, DARPA is exploring materials that can perform computation directly. Embedding compute within sensors and actuators could enable real-time decision-making with minimal latency, reduced power demands, and the ability to adapt continuously to changing conditions.

Together, these areas point toward a new class of systems where perception, decision, and action are tightly integrated at the hardware level. DARPA is not seeking incremental improvements or system-level concepts divorced from enabling hardware. Instead, the focus is on breakthroughs that could fundamentally reshape what robotic systems can do.

Additionally, while industry has emphasized human-like form factors designed to operate in human environments, of interest here are systems optimized for mission needs. Depending on the application, this could include designs that are smaller, larger, softer, or structurally unconventional, prioritizing performance and adaptability over familiarity. Responses to the RFI are due by May 27, 2026, at 2 p.

m. ET. Submissions will help inform future DARPA programs and guide the agenda for the upcoming workshop.

Participation in the workshop will be limited, with invitations extended to respondents whose ideas align with the agency’s technical interests and mission needs. Those selected may be asked to present their concepts and engage directly with DARPA program managers and peers across the research community. More broadly, this call underscores DARPA’s focus on the hardware foundations of autonomy.

Breakthroughs in this area could enable a new generation of systems capable of operating where today’s technologies fall short. Additional information on the RFI, including submission instructions, is available in Special Notice DARPA-SN-26-76. Published: April 27, 2026 Today’s robots are often limited by the need to sense, process, and act as separate steps.

– Julian McMorrow, MTO program manager