Rather than depending solely on digital processors, the approach enables robots to process information through their materials and structure. The framework centers on three methods (analog oscillators, physical reservoir computing (PRC), and algorithmic physical computing), each supporting tasks such as locomotion, sensing, and programmable control.
From Passive Bodies to Active Computing
Soft robotics has advanced rapidly in actuation, sensing, and power, but onboard computation has remained largely unchanged, still relying on rigid electronics. Earlier efforts used morphological computation, where a robot’s body mechanics simplify control. While useful, these approaches tend to be limited in flexibility and cannot easily be reprogrammed.
This study moves beyond that limitation by introducing physical computing as a structured process.
Instead of passive responses, the system explicitly encodes inputs, processes them through a physical “kernel,” and decodes outputs into actions. In doing so, the robot’s body becomes an active part of the computing system rather than just a mechanical support.
Three Paths to Embedded Intelligence
Building on this framework, the researchers outline three distinct approaches, each addressing a different aspect of physical computation.
Analog oscillators provide the simplest implementation. They convert constant inputs into rhythmic outputs, enabling coordinated movements such as walking or flapping. For instance, a pneumatic ring oscillator can generate phase-shifted signals that drive a quadruped’s gait. These systems are robust and integrate naturally into soft bodies, but their behavior is largely fixed, making them difficult to adapt without redesign.
Physical reservoir computing introduces more flexibility. Here, the robot’s body acts as a nonlinear system that transforms inputs into rich, high-dimensional signals. A trained readout layer then interprets these signals for specific tasks. This allows the same physical structure to handle multiple functions, from classifying payloads to adapting locomotion in real time. However, performance depends on careful calibration, as noise and variability can affect reliability.
Algorithmic physical computing takes a more digital-like approach. Using bistable mechanical elements, it enables logic operations and memory without electronics. Systems based on fluidic or mechanical circuits can execute programmed sequences, such as controlling robotic fingers or generating walking patterns. While this adds programmability, it also introduces challenges in terms of speed and scalability due to the limits of physical processes.
Balancing Capability and Constraints
Together, these approaches show that soft robots can perform increasingly complex tasks using embedded physical computation. Locomotion, sensing, and decision-making can all be handled, at least in part, by the robot’s own structure.
At the same time, each method comes with trade-offs. Analog systems lack flexibility, reservoir computing requires stability and noise management, and algorithmic systems remain slower and harder to miniaturize than electronic counterparts. These constraints highlight that physical computing is not a replacement for digital systems, but a complementary approach.
Looking ahead, the study points to three areas for development: physical memory, miniaturization, and hybrid systems.
Mechanical memory using bistable structures could enable nonvolatile data storage, while advances in microfabrication may allow more compact and capable physical circuits. Hybrid designs that combine mechanical computation with electronic readouts are likely to play a key role in scaling these systems.
Outlook
The framework provides a clear path toward more integrated soft robotic systems, where computation is no longer separate from the body. While current implementations remain at an early stage, they demonstrate that information processing can be embedded directly into materials and structures.
In practice, the most effective designs will likely combine physical and digital computing, using each where it is most effective. For applications tied closely to the body or operating in challenging environments, physical computing offers a compelling alternative.
Journal Reference
Wang, J., Zhou, Z., Kahak, A., & Li, S. (2026). Embodying physical computing into soft robots. Nature Communications, 17(1). DOI:10.1038/s41467-026-70866-6. https://www.nature.com/articles/s41467-026-70866-6
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.