Pinpoint Landings Mark New Development for UC Berkeley’s Salto Jumping Robot

Salto, UC Berkeley’s single-leg jumping robot, which has been in development since December 2016, makes progress by leaps and bounds every year. Duncan Haldane, who has long been involved in research on incredibly agile bioinspired robots, first debuted Salto when it could only manage two jumps in a row, although it already had quite impressive vertical jumping agility. Since then, Salto has been upgraded with a host of new abilities. Now, it can go on walks through Berkeley’s campus, navigate obstacle courses, and, in its latest achievement, it can stop jumping precisely where you want it to.

Until now, the small robot has not been great at landing, needing researchers to snatch it out of air mid-bounce to keep it from getting damaged. Justin Yim, the lead author on the ICRA paper that presented this development, says that Salto’s landing angle must be incredibly accurate, or the angular momentum transferred from the falling motion will be too much for the robot’s reaction wheel tail to negate. This is a challenge also faced by gymnasts: sticking the landing. The margin of error for Salto is about 2.3 degrees forward or backward. The reaction wheel tail works in a similar way to a person pinwheeling their arms to maintain balance, Yim says.

Prior work addressed stabilizing the landing and control of continuous jumping by setting the touchdown angle in mid-air, but greater precision is possible by using a liftoff angle to direct individual leaps. Reliable leaping and landing results from the combination of stance-phase balance control based on angular momentum, a launch trajectory that stabilizes the robot at the desired launch angle, and an approximate expression for selecting touchdown angle before landing.

A narrow foot also enables dynamic transitions between standing, hopping, and standing again. Previous versions of Salto had a larger foot than the small bar it is currently landing on, although the first design featured a point foot. The demonstration shows it with a small bar, which has since been replaced by a point foot with two more toes that reach the ground when Salto is crouched. This allows for the ease of control a point foot gives when bouncing, and the added benefit of a bar foot that applies roll torques when landing.

Motion capture is also a huge help when it comes to consistency, although Salto can now balance without it. The motion capture information sent to Salto are its velocity and some instructions for getting to the points marked out in a serious of precision jumps. And like always, there are so many exciting future directions Salto could take. Yim says he is most excited about doing things that robots haven’t been good at before, whether that be jumping on branches or mimicking other things animals can do that have challenged robots.