Branching Out in Soft Robotics Control

Soft robotic arms offer a revolutionary approach to manipulating objects and interacting with the environment. Unlike their rigid counterparts, soft robotic arms are constructed from flexible materials, such as elastomers or textiles, and are typically powered by pneumatic or hydraulic systems. This unique design allows them to bend, deform, and adapt to their surroundings, granting them a wide range of motion and dexterity.

The inherent flexibility and compliance of soft robots enable them to navigate through complex and confined spaces with ease. For instance, they can reach into tight crevices, handle delicate objects without causing damage, or even navigate through cluttered environments. This makes them particularly valuable in applications such as search and rescue missions, exploration in hazardous environments, or medical procedures where precision and adaptability are crucial.

It is these same properties that we value in soft robotic arms that make them challenging to control, however. The best way to develop an effective control mechanism that can deal with the flexibility and deformability of the materials involved is still an active area of research. Most present solutions rely on camera-based solutions which do work well enough, but they are not able to be used outside of a laboratory environment in diverse, real-world scenarios.

A team composed of researchers at the Sant'Anna School of Advanced Study and the National University of Singapore approached this problem from a completely different angle. Recognizing that plant life covers virtually all habitats on Earth, and that many plants are soft and flexible like soft robots, they devised a control system based on plant movement. They believed that by adopting similar principles of movement, they could design a soft robot capable of being controlled under nearly any conditions, not just under ideal laboratory conditions.

Contrary to popular belief, plants do move to accomplish specific purposes, like seeking sunlight or nutrients. But unlike the muscular system that is utilized by humans and animals to move, plant movement is governed by growth. This can be achieved by, for example, releasing hormones that cause the cells on one side of the stem to grow more quickly than on the other side. The researchers described this growth control process as operating something like a decentralized computing mechanism.

The controller was implemented on a 9-DoF modular cable-driven continuum arm. Three radially arranged actuators enable the arm to bend in six primary directions. Proximity sensors were embedded near the end-effector to provide information about the arm's location relative to a target. Behavior-based artificial intelligence tools, consisting of decentralized computing agents, were leveraged to simulate the controlled growth mechanism of plants.

The learning algorithm was trained to simulate two types of plant movement — circumnutation and phototropism. Circumnutation is a helical movement seen in many types of plants, while phototropism moves a plant in a particular direction to gather more sunlight.

These two types of movement are used initially in an exploration phase in which the robot gathers information about its surroundings. This is followed by a secondary reaching phase, in which the arm moves towards a predefined target to achieve a specific goal.

While this may seem like a somewhat simple architecture for a controller, it has proven to be effective. And importantly, this is the first control system for soft robotics to find success in real-world environments. The researchers note that their methods are applicable to any soft robotic arm with a similar actuation system, so this novel idea may power all manner of soft robots in the future.

At present, the team is working to extend the capabilities of their motion controller. In addition to reaching, they hope to also enable additional functionalities like target tracking and whole-arm twining.