In a world full of things that are much, much bigger than they are, insects manage to do pretty well with getting around. Some of the most successful insects are the social and cooperative ones, like ants, which can do incredible things such as using their bodies to create structures to get themselves across rivers.
In Ron Fearing’s lab at UC Berkeley, Carlos Casarez was inspired by behaviors like these to modify some VelociRoACHes to help each other climb up and over obstacles with the aid of an adorable little magnetic tether system.
Robots like UC Berkeley’s VelociRoACH (Velocity Robotic Autonomous Crawling Hexapod) are, relative to most other robots, very simple and inexpensive to manufacture, with high speed and good maneuverability. They’d be just the thing you might want to send a swarm of into disaster areas to search for survivors, except that they’re absolutely terrible at climbing over obstacles. The problem is that the underactuated nature that makes them so fast and cheap to produce also means that you can’t control the motion of each leg independently, which is what allows other legged hexapods (like RHex) to do parkour.
VelociRoACH isn’t going to be doing parkour anytime soon, but by getting a couple of them them to cooperate with each other like ants do, you can make a combined robot that’s much more versatile than either robot would be by itself, able to leap tall steps in a single bound! Or, nearly:
Casarez, lead author of an upcoming paper on this approach, told us how this all came together:
This is a novel design because depending on how a pair of VelociRoACH robots coordinate their actions, they can form a sort of modular robot with an extended jointed body and twice the number of actuated legs, or one robot can be an anchor while the other one uses the tether to provide pulling assistance. It’s also a minimal design because a single additional motor drives both the connection forming and the tether pulling assistance. My design solution combines previous approaches to robot cooperation to climb step or slope obstacles that I cite in the background section of the paper, which produces a more versatile platform than other systems.
The cooperating robots benefit from distributed actuation of modular robots to climb steps, while being able to break the modular robot up when it is too cumbersome to drag the connected robots up the step. The tether-assisted climbing mode fills in the last piece of the step climbing strategy— instead of just sending one member of the team up the step after climbing while connected, the winch can be used to pull the other team member up the step as well.
The tethering system itself consists of a rapidly-prototyped winch module with a polyethylene monofilament tether and magnetic connector that can be latched on to a compliant pin on the back of a second VelociRoACH.
Two VelociRoACH robots with a compliant connection. Left: Forming a connection begins with a magnetic connector tethered to the back robot. Right: Magnets engage to form a compliant pin connection between robots. Image: UC Berkeley
Once the tether is secure, the robots execute a series of “motion primitives” that together, if everything goes well, result in both robots on top of a 6.5 cm step that neither one would be able to climb by itself. Each primitive has a failure rate of about 50 percent for a variety of reasons, including the robots flipping over, twisting, getting stuck, or slipping off the step completely. This results in an overall success rate of 10 percent, which is not all that great, but Casarez explains how just some basic sensor feedback could significantly improve this:
While an overall success rate of 10 percent is too low to practically implement the cooperative step climbing strategy as is, it is still impressive considering that the robots used no sensory feedback to control each segment of the behavior. In the experiments performed in this paper, we simply prescribed a fixed gait and winch load control strategy for each phase of the behavior. In the future, we plan to improve the reliability of the cooperative step climbing behavior by adding closed-loop feedback control involving robot-to-robot localization, connection contact sensors, and IMU/motor torque information from each robot.
Once a single VelociRoACH robot encounters an obstacle it cannot climb (leading into primitive II), implementing robot-to-robot localization using vision or IR emitter/receiver pairs (emitters on one robot, receivers on the other) could assist the second robot with the winch module in finding the first robot and forming a connection with it. In addition, contact sensors could be used to sense whether the compliant connection is formed between robots (right before primitive III), or whether the tether is attached to the leading robot (right before primitive V) before continuing on with the next phase of the behavior.
Cooperative insects like ants are able to cooperatively form more complex structures as their numbers increase, so we also asked Carlos about whether his future plans included adding more VelociRoACH robots into the mix:
If we think about scaling the step-climbing behavior by adding more robots with winch modules and connection components to form a chain of 3-10 robots, we can infer that the increasingly long chain of robots with more joints and actuators will be able to climb over taller and taller steps. In addition, a long chain of robots could potentially cross gaps in terrain that are smaller than the length of the robot chain. The chain of robots could wedge themselves in the gap and boost the front robot in the chain past the opposite side of the gap. However, as we add more and more robots to these cooperative behaviors, it becomes harder to reliably perform the behavior—if the connection between one pair of robots in the chain fails, then the entire operation fails.
I think there is some more interesting work in exploring how far two-robot cooperation can go. For example, instead of climbing a step, the robot with the winch could be used as an anchor for a tethered VelociRoACH that explores down an unknown chasm, which can then be retrieved after exploring. Also, if you want to get 10 VelociRoACH robots over a two-robot climbable obstacle, you can simply follow the cooperation primitives in the paper, then keep attaching more winch robots to the back of the leading winch robot. The first robot that gets over the obstacle is the anchor, and tether-assisted climbing can be used to pull as many robots as you want up the step.
I am also very interested in using the components presented in this paper to enable cooperation between heterogeneous robots. Let’s forget about step climbing altogether and attach the tethered magnetic connector of a VelociRoACH with winch module to a quadrotor, which can fly over whatever obstacles are in the way and then anchor itself to the environment. The VelociRoACH with winch can then pull itself over whatever terrain is in between it and the quadrotor.
We’ve been fans of heterogeneous robot swarms for a long time, and it’s very cool to think about how teams of VelociRoACH robots might be able to work with teams of quadrotors like this. The other thing to reiterate is how simple and cheap these roachbots are, and if you can use a bunch of them to mimic (or even improve upon) the functionality of larger and more expensive robots, you could save a lot of time and resources by relying on them instead. Imagine sending a swarm of these little guys into, say, Fukushima to gather data about the environment inside: you know that half of them are going to get fried by radiation, and none of them are going to come back, but it just isn’t that big of a deal, because they’re essentially expendable. And as long as they can cooperate with each other to get up stairs and over rubble that’s bigger than they are, there will be no stopping them. Even if you want to.
“Step Climbing Cooperation Primitives for Legged Robots With a Reversible Connection,” by Carlos S. Casarez and Ronald S. Fearing from UC Berkeley, will be presented later this month at ICRA 2016 in Stockholm, Sweden.
[ UC Berkeley ]