UC Berkeley's "Hyper-Aggressive Pogo-Stick" Robot Now Works Outdoors

Salto gets an upgrade and now it can go for a bounce outside the lab

3 min read
UC Berkeley jumping robot Salto
New version of Salto with upgraded controller can now go for a bounce outside the lab.
Image: UC Berkeley via YouTube

UC Berkeley’s Salto has been one of our favorite robots since 2016, which makes it old-ish in robot years. While it’s kept the same “hyper-aggressive pogo-stick” concept, a series of upgrades has endowed Salto with with the ability to do increasingly more dynamic maneuvers.

The original Salto could make two jumps in a row. Thrusters added in 2017 gave the robot the control it needed to chain together multiple jumps. And late last year at IROS, an improved controller gave Salto the intelligence that it needed to make pinpoint jumps that allowed it traverse a series of vertical obstacles (and more).

The big constraint has always been that in order for Salto to keep itself upright and in one piece, it had to jump within a motion-capture environment, which limited its usefulness to (let’s be honest) not much more than a cool research project and highly effective YouTube video view generator.

Today at ICRA, UC Berkeley roboticists Justin Yim and Eric Wang (from Ron Fearing’s Biomimetic Millisystems Lab) presented the latest version of Salto, which adds the sensing and computing required to do away with the motion-capture system completely. Salto can now jump as much as you want out of the lab, and in fact completely outdoors.

Salto is not jumping autonomously—Justin is “driving” the robot, keeping it on the sidewalk and avoiding obstacles since Salto doesn’t have any onboard sensing. Jumping up onto that step was human-driven as well, and the description of the sequence of motions in the paper reminds us why taking humans out of the loop is such a good idea:

The human operator directed the robot to jump in place, and then advance in a straight line towards and onto a step 0.43 m high. The robot was then directed to hop to the right on top of the step, and then towards the left. This run lasted 19 seconds total before human error directed the robot into a bush on the left.

Oops.

Salt uses thrusters and a tail Salto features a series-elastic power modulating leg. In the air, a balanced inertial tail and two small lateral thrusters control the robot’s attitude. Image: UC Berkeley

The new onboard attitude estimation and hopping control system (called SHOVE, for SLIP Hopping Orientation and Velocity Estimator) is robust enough that Salto can jump continuously even on compliant surfaces like foam. It uses dead reckoning to estimate how much its position changes with each jump, and even after 300 jumps over the course of several minutes, this estimate only drifts by a meter or two (representing an error of less than 1 centimeter per jump). 

However, the tricky part right now is accurately estimating the Salto’s attitude, by which we assume the researchers mean its orientation (as opposed to how “hyper-aggressive” it is). Attitude estimate errors of about 1 degree result in foot placement locations that vary by about half a meter from jump to jump, and in practice, this means that Salto can’t plan its jumps with enough precision to reliably climb stairs.

The new onboard attitude estimation and hopping control system uses dead reckoning to estimate how much its position changes with each jump

This is being worked on, though, and the researchers expect that “higher precision estimation and control can enable jumping on more finely varied surfaces like stairs, furniture, or other outcroppings” as well as “soft substrates like upholstery or natural foliage.”

The researchers tell us that Salto’s hardware is capable enough at this point that aside from potentially upgrading the motor or battery for more jumping power or run time, the focus now will be on new behaviors, although they’re toying with the idea of adding some kind of gripping foot so that Salto can launch from, and land on, tree branches (!). And as for that “pair of legs” and “arms” that Professor Ron Fearing mentions in the video, that would be quite an upgrade as well, but I’ll always have a soft spot for Salto as a bouncy little monoped.

“Drift-free Roll and Pitch Estimation for High-acceleration Hopping,” by Justin K. Yim, Eric K. Wang, and Ronald S. Fearing from UC Berkeley, was presented at ICRA 2019 in Montreal, Canada.

[ UC Berkeley ]

The Conversation (0)

How the U.S. Army Is Turning Robots Into Team Players

Engineers battle the limits of deep learning for battlefield bots

11 min read
Robot with threads near a fallen branch

RoMan, the Army Research Laboratory's robotic manipulator, considers the best way to grasp and move a tree branch at the Adelphi Laboratory Center, in Maryland.

Evan Ackerman
LightGreen

“I should probably not be standing this close," I think to myself, as the robot slowly approaches a large tree branch on the floor in front of me. It's not the size of the branch that makes me nervous—it's that the robot is operating autonomously, and that while I know what it's supposed to do, I'm not entirely sure what it will do. If everything works the way the roboticists at the U.S. Army Research Laboratory (ARL) in Adelphi, Md., expect, the robot will identify the branch, grasp it, and drag it out of the way. These folks know what they're doing, but I've spent enough time around robots that I take a small step backwards anyway.

This article is part of our special report on AI, “The Great AI Reckoning.”

The robot, named RoMan, for Robotic Manipulator, is about the size of a large lawn mower, with a tracked base that helps it handle most kinds of terrain. At the front, it has a squat torso equipped with cameras and depth sensors, as well as a pair of arms that were harvested from a prototype disaster-response robot originally developed at NASA's Jet Propulsion Laboratory for a DARPA robotics competition. RoMan's job today is roadway clearing, a multistep task that ARL wants the robot to complete as autonomously as possible. Instead of instructing the robot to grasp specific objects in specific ways and move them to specific places, the operators tell RoMan to "go clear a path." It's then up to the robot to make all the decisions necessary to achieve that objective.

Keep Reading ↓ Show less