Professor Hadas Kress-Gazit has a kind of “hands-off” approach when it comes to robotics.

Kress-Gazit wants to give a robot a goal – find the green object, for example – and then let the robot decide, on its own, how to accomplish the task.

How can a robot have so many movement and behavior options to choose from exactly?

Kress-Gazit and her fellow researchers at Cornell University and the University of Pennsylvania are developing modular robots – ones that change their shape based on a given task or given surroundings.

“This is the first time modular robots have been demonstrated with autonomous reconfiguration and behavior that is perception-driven,” said Kress-Gazit  , an associate professor at Cornell. Research results were published on Oct. 31 in Science Robotics.

The robots feature wheeled, cube-shaped modules that magnetically detach and reattach to form new shapes with different capabilities.

The interchangeable modules, developed at the University of Pennsylvania, are connected to a sensor module. Additionally, the modules are equipped with multiple cameras and a small computer for collecting and processing data about its environment.

A high-level planner in the software directs the robot’s actions, reconfiguration, as well as perception algorithms that can map an environment.

Once the robot is given a task, the planner searches an open-source library consisting of 57 possible robot configurations and 97 behaviors, including “pickUp,” “highReach,” drive, or drop.

Perhaps “Proboscis,” where a long arm moves to the front of the robot, is required. Or maybe the modules need to be arranged in “Scorpion” mode, with perpendicular lines and a horizontal row in front.

In the first of three experiments, a robot was instructed to find and deliver all pink and green objects to a designated zone marked in blue. The robot used the “Car” configuration to explore the area, and then morphed itself into “Proboscis” to retrieve the target from a narrow pathway.

Although the effort took 24 tries, a second experiment demonstrated that a modular robot could autonomously place a circuit board in a mailbox – one that was marked with pink tape and fixed at the top of a staircase.

The Cornell and UPenn researchers found that the hardware and low-level software were the technology components most prone to error, especially when dealing with more complex obstacles like a set of steps.

If such issues are resolved, however, Kress-Gazit said the robots could someday support search-and-rescue applications, especially in environments with uncertain and unpredictable terrain.

Kress-Gazit told Tech Briefs, via email, about the “great promise” of modular robots, especially when they have greater autonomy. The professor’s edited responses are below.

Tech Briefs: What does “modular” mean exactly?

Professor Hadas Kress-Gazit: It means that the robots are created by connecting different pieces, or modules, into larger shapes. In the images and videos, each white cube is a module.

Tech Briefs: Compared to the robots out there today, what is special about your robot?

Prof. Kress-Gazit: The modular robot can be connected to create different shapes. Some examples for robot shapes: We can create a snake robot (modules connected in a line), a rolling wheel robot (a “snake” where we attach the last module to the first) and a quadruped (modules connected to form 4 legs, and then connected to a few modules that form a body) .

Tech Briefs: How are the shapes made?

Prof. Kress-Gazit: Our robot is created by connecting cube faces. Any shape that can be built by putting cubes next to each other can be built. The robot chooses which shape to use based on the task it is trying to perform and the environment in which it is operating.

There are three key technological pieces to our work:

  1. Hardware design, which is the technology that enables the building of the modules and the reversible connections between them (based on electro-permanent magnets).
  2. Perception algorithms that, given the sensor information (camera images, depth information), reason about what the environment looks like and whether the robot can perform its task in the environment.
  3. A high-level planner that, given a task, decides on the shape and control of the robot.

Tech Briefs: What are the autonomous decisions that the robot is making? What is the robot “deciding” exactly?

Prof. Kress-Gazit: The robot takes in information from its sensors and, based on that and the overall task that we gave it, decides how to act. The actions include motion (where to move, how to control the motors to create the desired movement), which is a decision all autonomous robots make, as well as reconfiguration – how to reshape its own body. The reconfiguration actions are unique to modular robots.

Tech Briefs: How is the modular robot able to perceive its surroundings to make these decisions?

Prof. Kress-Gazit: The robot has an onboard RGBD camera that provides images and depth information, and a camera that is pointed at the robot's body to allow for autonomous reconfiguration.

Tech Briefs: What applications do you envision with this kind of robot?

Prof. Kress-Gazit: The great promise of autonomous modular robots is their flexibility and ability to change shape. Applications that require the robot to adjust its shape and behavior based on the environment are especially suitable for this type of robot, and they include disaster relief and exploration of challenging environments.

A modular robot, in one of many possible configurations (Image Credit: University of Pennsylvania/ModLab/Tarik Tosun)

What do you think about the possibilities of modular robots? Share your comments and questions below.



Transcript

00:00:00 we present an integrated system for perception driven autonomy with modular robots this video showcases three Hardware experiments that demonstrate the capabilities of our system experiment one takes place in the l-shaped office environment shown here the robot has no knowledge of the environment before the experiment begins its task is to explore find all objects

00:00:20 of interest and bring them to a designated drop-off zone the following video was shot in one take using multiple cameras the robot begins exploring the unknown environment it quickly recognizes the Blue Square that marks the drop-off zone and remembers its location the robot continuously performs slam building up a 3d map of its environment as it drives it selects

00:00:39 waypoints for exploration using a next best view algorithm that maximizes information gain about the environment while also avoiding obstacles the robot recognizes that the pink object is positioned in a narrow gap identifying this as a tunnel type environment it autonomously determines that a different configuration is needed to reach the object the robot begins the

00:00:58 reconfiguration sequence by lying down so that modules that need to move will have their wheels in contact with the ground a module detaches and uses its wheels to drive to the front of the car where it spins in place to establish perfect alignment then drives into place and attaches localization during reconfiguration is provided by a downward-facing RGB camera

00:01:21 mounted high up on the sensor module this is used to track April tagged markers attached to the top of each reconfiguring module the robot has successfully reconfigured into their proboscis configuration now it's able to drive into the tunnel and retrieve the object because the proboscis is unable to turn the robot must reconfigure back into its original

00:01:42 configuration in order to drive the object to the drop-off zone it starts by dropping the object and backing up to make space to reconfigure it then lies down again and begins the reconfiguration sequence to make reconfiguration more reliable we've developed techniques to assure strong connection between the modules for example when modules connect the driving

00:02:01 module over drives a stock point to ensure the magnets make good contact the robot picks up the pink object and begins navigating to the drop-off zone as it drives it spots another object which will be its next target when it thinks it's in the right place it confirms that it can see the Blue Square and drops off the pink object it then begins navigating to the other

00:02:30 object it spotted the environment characterization system recognizes that there's enough space for the robot to pick up the green object without reconfiguring the robot drives in and latches to the green object using its magnets the object becomes caught in a crack in the floor surface and is freed by a human operator note that this is a

00:02:52 defect in the experimental setup which is assumed to have a floor with no cracks with the green object attached the robot once again makes its way back to the drop-off zone when it believes it's reached the drop-off zone the robot once again performs a search for the blue square to ensure that it's in the right place

00:03:23 once it finds it the robot drops the object off completing its mission in the next two experiments the robot will help researchers at the University of Pennsylvania mail a circuit board to collaborators at Cornell first the robot needs to place a circuit board in a mail box the robot begins in the Skorpion configuration holding the circuit board after exploring it identifies the

00:03:46 mailbox which is marked with a pink rectangle it characterizes the environment in front of the mailbox as stairs and moves into position to reconfigure by popping off its front two wheels the scorpion transforms into the snake configuration which is able to climb the stairs at the top it drops the circuit board

00:04:10 into the mailbox and descends at the bottom the front wheels reattach to form the Scorpion configuration again next the robot needs to place a postage stamp on the box for shipping the robot begins in the car configuration holding the postage stamp on its right wheel after navigating around an obstacle it locates the pink square on the mailbox where it needs to place the postage

00:04:42 stamp it recognizes that the target is too high for the car configuration to reach so it chooses to reconfigure into the proboscis you the proboscis reaches up and places the stamp on the box completing the mission thank you for watching