Purdue Helps Robots Navigate Unstable Ground

Purdue University researchers are stepping in to teach robots how to navigate the swaying deck of a boat or the uncertainty of a moving walkway — enabling their practical use in more real-world situations. Watch this video to see robots find their sea legs.

“These robots are used for inspection, maintenance, disaster response, surveillance, security, and many other applications,” said Yan Gu  , associate professor of mechanical engineering. “But there remains a major challenge to overcome for legged locomotion, and that is reliable operation on a moving surface.”



Transcript

00:00:00 In the real world, we can use legged robots  for inspection, maintenance, firefighting,   disaster response, emergency response and  surveillance security. There are many applications   of legged robots. We know Boston Dynamics has  been constantly releasing very impressive videos   of legged robots. Their robots can do a back flip  and jump to high surfaces. These robots, they have   amazing performance capabilities when addressing  stationary surfaces. But once this floor begins   to move, we see that their performance can  deteriorate really fast. There remains a   major challenge for legged locomotion which is  to enable reliable operation on a moving surface:   for example, ships, aircraft, trains, commuter  rails. Our approach is to explicitly consider   the movement of the ground and incorporate the  movement into the dynamics model of the robot.   So the robot has about 30 joints and they  have various sensors on it. On top of it,  

00:01:03 it has a LIDAR sensor, and in front of  the robot there's a depth sensor which   allows the robot to see the terrain right in  front of it. The treadmill is very capable:   it can perform both rocking motion as well  as a sway motion. Also has a split belt,   so the two belts can be actuated separately,  and we can change their speed separately.   Also has force plates, which can measure  the ground reaction force on the robot,   and it's also integrated with the motion capture  cameras. We have 20 motion capture cameras that   can capture the true movement of the robot while  it walks on the moving surface, so that we can   use the movement data to analyze how well our  algorithm can enable reliable locomotion. Right   now, we are just taking baby steps. We have  derived new models, new estimation algorithms,   and new control approaches that can make  a robot walk reliably on a moving surface.