The Soft Robot

For most preliminary five-legged designs, a sequence of inflating back legs, inflating front legs, deflating back legs, and deflating front legs resulted in an undulating gait when deflate times were allowed to be longer than inflate times. By inflating sharply and deflating slowly, the contact points between the legs and the ground moved forward quickly but did not move back, moving the center of mass of the robot in a controllable direction. However, the resulting movement was slow, since the center of the robot was always in contact with the ground. This constraint also meant that the five-legged design could not raise itself above an object to grasp.

Additionally, keeping one leg permanently deflated (acting as a pivot) and inflating the remaining legs in sequence resulted in a turning gait. However, the angle change was minimal, once again due to most of the soft robot remaining in contact with the ground.

Quadruped Robot Designs

To mitigate the aforementioned issues with the five-legged design, four-legged designs were adopted. Each leg contained two air channels, allowing a paddling motion for each leg to be developed. This was achieved by inflating the back channel (pushing the leg down and forward), inflating the front channel (pushing the leg further down and back), deflating the back channel (pulling the leg up and back), and deflating the front channel (pulling the leg up and forward). By actuating diametrically opposed legs synchronously, interleaving actuation sequences of opposing pairs of legs, and, once again, allowing deflate times to be longer than inflate times, the soft robot was able to move along one of its axes of symmetry.

The quadruped is pictured below, along with a diagram showing the internal structure. For our project, the conceptual design and dimensions in the diagram were used in our design.  In our design, we require a place for the manipulator to grasp the soft robot. Our initial design features a silicone tab attached to the top of the soft robot. This tab is visible in the picture below. Other differences are explained in the Fabrication section. 

Diagram of the quadruped design, Stokes, et. al. 2014.

Designing for Manipulation and Vision-Based Tracking

Because the silicone tab did not stand upright, a 3-D printed tab was designed instead and mounted to the soft robot, as shown below. The tab was colored yellow and the soft robot was given the same color in order to produce a ensure that the youBot's perception system can reliably track the soft robot when placed at a reasonable distance away from the youBot and and be robust to changes in the soft robot's orientation.

After testing the soft robot, it was concluded that the yellow tab alone was sufficient to track the soft robot's location using the vision system. Emphasis was placed on coloring the robot such that its orientation with respect to the youBot in any of the four cardinal direction is easily distinguishable. That way, the youBot can provide appropriate commands to move the quadruped toward its goal without resorting to a "guess-and-check" type of control strategy.

The design concept operates on the same principle as a binary color encoder with resolution down to the nearest quadrant. We describe its operation through an example. Assuming the two black legs represent the front of the robot, if a black leg is visible to the left and a white leg to the right in the camera image, then we know the front of the robot is facing left in the camera image.  Now, assuming that we want the soft robot to move left in the camera image, it is required to actuate a gait to move the robot forward. By color coding the robot in this way, the control system is robust to variations in how the soft robot is oriented when placed on the ground.

To prevent leaks and having the tubes from being damaged or removed when the youBot manipulator picked up the soft robot, the piping was glued into the soft robot. Because this interfered with the plastic tab, yellow tape was instead used to secure the tubing. The tape served three purposes: to keep the tubing from separating, to act as a handle for the manipulator to pick up the robot, and to act as a color "blob" that could be seen by the youBot vision system.

The tether length was an important factor in the design; the considerations were range of motion, potential for entanglement with the youBot manipulator and wheels, effect of the tether's drag on quadruped mobility, and dynamic range of the vision sensor used for tracking the soft robot.

The published range of the depth readings for the ASUS Xtion Pro camera is between 80 cm and 350 cm. Because the manifold was situated toward the rear of the youBot, 50 cm was added to the required forward travel of the soft robot beyond the front of the youBot. To satisfy the camera constraints, a tether length between 120 cm and 400 cm was required. We ultimately chose 150 cm as the tether length, as this satisfied having reasonable mobility within the field of view of the sensor (between zero and 100 cm from the front of the robot), while keeping the effect of drag small and risk of entanglement low.