Latest upgrade to Salto allows it to land precisely on its target

4 Jun 2020

Salto is now capable of landing precisely where you want it. (Image Credit: UC Berkeley)

Since it’s development, we’ve seen what UC Berkeley’s Salto robot is capable of, from hopping off walls to running to jumping over obstacles, etc. Each year, the robot just seems to get better and better with improved capabilities. The tiny one foot long leaping robot’s latest upgrade is quite impressive. Presented at Virtual ICRA 2020, Salto has recently learned how to stop jumping at the exact spot you want it to.

Until now, the only way to get Salto to stop jumping without damaging or destroying itself was by having someone with perfect timing grab it out of the air while it’s descending. With the latest improvements, Salto can land perfectly while maintaining its balance, which has been an issue in the past. According to Justin Yim, Salto’s landing angle has to be highly accurate, or else the angular momentum will be too much for the reaction wheel tail to negate. If the robot is to land successfully, Salto will need about 2.3 degrees of room forwards or backward.

What’s even more remarkable is that Salto is landing on just one tiny foot. In the video, you can watch the robot land on its one-dimensional bar foot. According to the ICRA paper, it can also land on just a point foot, even though its thrusters have a difficult time keeping it upright. The robot does most of the landing work itself while it’s in mid-air. To successfully land, the robot uses the same hardware as the previous version, consisting of a spinning inertial tail along with two propeller thrusters designed for multi-axis stability.

The landing feature isn’t the only thing that’s new with the robot. It also has improved control over where it’s heading, allowing it to set a target for its next jump to a specific area. This new capability comes from both precise leaping to targets and balanced landing. This allows Salto to have improved control over its next leap since it has the opportunity to land and precisely choose its next target, as opposed to when it bounces. Salto has also significantly reduced its landing target’s standard deviation from 9 centimeters down to 1.6cm. This basically means that Salto will be able to handle much smaller targets like ledges and tree branches.

Yim says he is excited about Salto’s potential and what it may be capable of in the future, which also includes doing tasks that other robots haven’t been able to perform well in. Such tasks include jumping on branches or imitating other things that animals are capable of doing, but robots haven’t been able to follow.