MARLO vs Wave Field: Round 2 (Blindfolded) [Long Version]

MARLO is one of three ATRIAS 2.1 bipedal robots designed and built by Prof. Jonathan Hurst and the Dynamic Robotics Laboratory at Oregon State University.

Experiment run on Saturday, 23 July 2016. Navigation is done via the Xbox controller. Right now, MARLO is still walking without the benefit of a camera. The cable dragging behind the robot is a power cord; our LiPo batteries had gone bad, and, due to a thunderstorm two nights before this experiment, the hobby shop where we normally buy them was without power, and hence closed! Kind of ironic. The cable is a distraction, for sure, but hey, it is not helping the robot to keep itself upright.

The feedback controller being run is the exact same one as in this simulation we had posted a few days earlier: https://youtu.be/TpsJ3jPVqGA What is different from the controller used in our previous attempt https://youtu.be/YErF0cyPI-g ? We completely re-designed the algorithm that determines the lateral (side-to-side) leg placement. Details will be published in the next few months, but roughly speaking, with each impact of the robot’s feet on the ground, we use the angles of MARLO’s joints and knowledge of the lengths of the various links in its legs to form an estimate of changes in ground height. The local estimate of ground profile is used to determine how high its feet need to be lifted in those side-step maneuvers in order to place the foot at its desired location, without banging the foot into the ground prematurely, as was happening before.

This is all well and good, but for the moment, the algorithm has been hand coded for MARLO, which is the antithesis of what our research is about. We want general principles that can be applied to other robots and lower-limb prostheses, such as these:
DURUS https://www.youtube.com/watch?v=1fC7b2LjVW4 and Vanderbilt Leg https://www.youtube.com/watch?v=sl1IXs0j4Ww . With the new academic year, we will take what we learned on MARLO and improve our mathematics so that the algorithm used here…or better yet…an improved version of it, follows from clean, easily transferable design principles.

In case you are wondering, we develop our software in MATLAB and then port it to the embedded controller via Real-time Simulink, previously called XPC-Target.

We wish to thank Prof Jonathan Hurst, Mikhail Jones, Andy Abate, Ryan Domres and the entire team at Oregon State University for providing us with much needed upgrades to the ATRIAS hardware. We also thank Jerry Brusher, Mathworks, for his help with Real-time Simulink code that runs without requiring access to a network.

You can find our publications on bipedal robots here: http://web.eecs.umich.edu/faculty/grizzle/publications-robotics.html

Видео MARLO vs Wave Field: Round 2 (Blindfolded) [Long Version] канала Michigan Robotics: Dynamic Legged Locomotion Lab