American Scientist

Scientists have been trying to build snakelike, limbless robots for decades. These robots could come in handy in search-and-rescue situations, where they could navigate collapsed buildings to find and assist survivors. With slender, flexible bodies, limbless robots could readily move through confined and cluttered spaces such as debris fields that are inaccessible to or too dangerous for human rescuers, but where walking or wheeled robots tend to fail.

However, even the most advanced limbless robots have not come close to moving with the agility and versatility in difficult terrain of worms and snakes. Even the tiny nematode worm Caenorhabditis elegans, which has a relatively simple nervous system, can navigate through difficult physical environments.

As part of a team of engineers, roboticists, and physicists from the Georgia Institute of Technology led by Daniel I. Goldman and Lu Hang, we wanted to explore this discrepancy in performance. Most attempts to create wormlike robots have used a neuroscience approach in an attempt to recreate how animals sense and react to obstacles. Instead of looking to neuroscience for a solution, our team turned to biomechanics to build a robot model that drove its body using forces similar to how worms and snakes power their movement.

Biomechanical Design

Organisms have evolved intricate nervous systems that allow them to sense their physical surroundings, process that information, and execute precise body movements to navigate around obstacles.

In robotics, engineers might design algorithms that behave like neural systems. The algorithms process information from sensors on the robot’s body—a type of robotic nervous system—and use that information to decide how to move. These algorithms and systems are usually complex.

Our team wanted to figure out a way to simplify these systems by highlighting mechanically controlled approaches to dealing with obstacles that don’t require sensors or computation. To do that, we turned to examples from biology.

Animals don’t rely solely on their neurons (brain cells and peripheral nerves) to control movement. They also use the physical properties of their bodies—for example, the elasticity of their muscles—to help them react to their environment spontaneously, before their neurons even have a chance to respond.

For a robot to complete the same task, scientists can either design an algorithm that relies upon sensors, or they can carefully design a physical system that automatically reacts to obstacles.

• Facebook
• Twitter

Whereas computational systems such as algorithms are governed by the laws of mathematics, mechanical systems are governed by physics. For a robot to achieve the same task, either scientists can design an algorithm that relies upon sensors, or they can carefully design a physical system that automatically reacts to obstacles.

For example, limbless robots and animals move through the world by bending sections of their bodies left and right, a type of movement called undulation. If they collide with an obstacle, they have to turn away and go around it by bending more to one side than the other.

Scientists could build a robot that can handle obstacles by attaching sensors to its head or body. They could then design an algorithm that tells the robot to turn away or wind around an obstacle that it “sees” or when it “feels” a large enough force on its head or body.

Alternatively, scientists could carefully select the robot’s materials and the arrangement and strength of its motors so that collisions would spontaneously produce a body shape that turns to avoid an obstacle without the use of sensors. This robot would have what scientists call mechanical intelligence.

Legged and aerial robots often incorporate both strategies of robot development; however, less is known about the forces behind undulation, so most limbless robots have used the algorithmic approach. If researchers can understand the biomechanics of how undulating organisms’ bodies respond to contact with objects in their environment, they can design better robots that can deal with obstacles without relying solely on complex algorithms.

Designing Mechanical Intelligence

If you compare a diverse set of undulating organisms with the increasingly large zoo of robotic “snakes,” one difference between the robots and biological undulators stands out: Nearly all undulatory robots bend their bodies using a series of connected segments with motors at each joint. In contrast, all limbless organisms, from large snakes to microscopic nematodes, bend not from a single, rotational joint–motor system but instead through two bands of muscles on either side of the body. To an engineer, this design initially seems counterintuitive. Why control something with two muscles or motors when one could do the job?

To get to the bottom of this question, our team built a new robot called MILR, for Mechanically Intelligent Limbless Robot, inspired by the two bands of muscle on snakes and worms. MILR has two independently controlled cables that pull each joint left and right, bilaterally.

We found that this method allows the robot to spontaneously move around obstacles without having to sense its surroundings and actively change its body posture to comply with the environment.

Rather than mimicking the detailed muscular anatomy of a particular organism, MILR applies forces to either side of its body by spooling and unspooling a cable. This system mirrors the muscle activation methods that snakes and nematodes use, in which the left and right sides take turns activating. The body turns by tightening the cables on one side while the cables on the other side relax and are pulled along passively.

By changing the amount of slack in the cables, we can achieve varying degrees of body stiffness. When the robot collides with an obstacle, it selectively maintains its shape or bends under the force of the obstacle, depending on the cable tension, and whether the obstacle struck on the activated side.

Head-on collisions that would stop or jam a neuroscience-based robot instead naturally led to a redirection around the obstacle for the mechanically intelligent robot.

• Facebook
• Twitter

We found that if the robot was actively bending to one side and it experienced a force in the same direction, the body complied with the force and bent further. If, alternatively, the robot experienced a force that opposed the bend, it would remain rigid and push itself off the obstacle.

Because of the pattern of the tension along the body, head-on collisions that would stop a simply controlled, serially connected robot instead naturally led to a redirection around the obstacle for the mechanically intelligent robot. MILR could push itself forward consistently.

To investigate the benefits of mechanical intelligence, we built tiny obstacle courses and sent nematode worms through them to see how well they performed. We sent MILR through a similar course and compared the results.

MILR moved through its course about as effectively as the real worms. When they collided with obstacles, we noticed that the worms and MILR responded with the same types of body movements.

The principles of mechanical intelligence extend beyond the realm of nematodes. Future research could look at designing robots based on a host of other types of organisms for applications ranging from search and rescue to exploring other planets.

This article is adapted from a version previously published on The Conversation (theconversation.com).