As a general rule, traditional robots made from metal frames, circuit boards, and rigid sensors can be extremely precise in their movements, while soft robots made from flexible materials like silicone or hydrogels move with far less precision. Despite this disparity in performance, soft robots are still more desirable than their rigid counterparts for applications that require a delicate touch, such as interacting closely with humans or handling fragile objects.
But there is another application area where no existing robotic technology is able to fully meet our needs. These are applications in which a gentle touch is required in conjunction with a high degree of precision in movement. Engineering a system with this combination of traits is exceedingly difficult, yet for the next generation of robot-assisted surgery platforms and medical devices, these are exactly the capabilities that will be necessary.
A look at the arm (📷: Rice Advanced Materials Institute)
We may not be as far from that goal as it seems, however, thanks to the work of a team of researchers at Rice University. They have developed a soft robotic arm that can be moved with precision via an optical control system to handle even the most delicate of objects. In addition to making advances in materials science, the team also developed an artificial intelligence (AI)-powered controller to move the arm exactly where it needs to go.
The arm is made from a special light-responsive polymer known as azobenzene liquid crystal elastomer, which bends or contracts when exposed to specific wavelengths of visible light. A blue laser beam projected onto the material causes it to shrink and bend in the direction of the beam, similar to how a plant stem bends toward sunlight. What makes this material especially useful is its ability to relax quickly when the light is removed, enabling real-time motion control.
Unlike most robots, which are tethered to power sources or burdened with onboard electronics, this soft robotic arm is completely wireless and untethered. All motion is induced remotely by shaping the laser light using a spatial light modulator, which splits the beam into many smaller beamlets. Each beamlet targets a different point on the robotic arm, effectively creating independent, controllable joints with a theoretically infinite number of degrees of freedom.
To make control of the system intuitive and precise, the researchers trained a convolutional neural network to predict the specific light pattern required to create a desired movement. The team ran a series of training experiments, illuminating the arm with various beamlet patterns and recording the resulting configurations. From this dataset, the AI learned how to generate exact patterns for precise movements, such as curling around an obstacle or reaching for a target.
Currently, the prototype only moves in two dimensions, but future versions are expected to bend and operate in full three-dimensional space. With further improvements, such systems could find use in biomedical devices, minimally invasive surgical tools, or even industrial applications that require manipulating fragile materials.
