1. Quantum Gate
Join Date
Aug 2014
Posts
99
Rep Power
19

## Biped Gait Algorithms

Been studying the DARwIn-Op walking code and have found several papers on the subject, so I'm posting them here in hopes that others will find them as useful as I did.

1) "Gait Pattern Generation and Stabilization for Humaniod Robot Based on Coupled Oscillators"

2) "DARwIn-OP Humaniod Robot Kinematics"

3) "Mobile Robot Modeling, Simulation And Programming", Appendix D Walking Parameters

2. ## Re: Biped Gait Algorithms

Since I've still not been very productive (and don't have any hardware to test it on even if I had been more productive), I might as well add what was supposed to eventually become my master's thesis if not for my brain continuing to crap out and having to drop out of grad school. Basically, it is a heuristic search for gaits for arbitrarily designed robots using prismatic or rotational actuators. The offline search produces the primary gaits for the robot, and those are also used as seeds for the online search when the robot parameters are altered by local disturbances (changing the mass distribution in the robot by moving the arms and/or carrying a weight). In addition to collision checking, the fitness function of the search does a bit of basic statics analysis of the full robot at multiple timepoints (a multiple of the number of points used to describe the spline) in the gait spline using each actuator position, link length, link COG, and link mass.

I pretty much wanted to make a system that could somewhat learn how to walk on its own. If a weight is added to the back of a biped robot and causes it to exceed the IMU limits expected by the gait (leans too far backward, swings too far side to side, and/or falls down), then it stops and reposes itself to try to find its new properties. It first uses some joint offsets to try to stabilize the gait, and if that fails it moves onto a quick gait search seeded by the splines of its last gait and its new estimate of bot properties. Lather, rinse, repeat until it gets a gait producing stable movement at the expense of the speed of, and/or distance traveled by, each footstep cycle.

Not much of an idea how well it would actually work with hardware.

The mostly static parameters of the search are the same as can be used to describe a robot and its actuators in ROS: a URDF with every link's dimensions plus the limits in each direction of the position, speed, and effort of each joint (with hard and soft limits on each; hard limits will rarely/never change, but soft limits can be used to tune the resultant gaits of the search). The remaining inputs to the search are the primary tunable parameters: 1) the distance traversed with each step; 2) the length of time to complete each cycle of the gait; 3) the number of points used to produce the cyclic spline describing each actuator's movement profile over a gait cycle; 4) the 'injury' modifier of each joint in both directions (simple multiplier to force the search to minimize the movement of key joints that are prone to heavy power usage, overheating, or other hardware difficulties that have resulted in repeated replacement of the servos in those joints); 5) the limits in each direction of X/Y/Z translation and R/P/Y rotation of each IMU.

There are currently 1 users browsing this thread. (0 members and 1 guests)

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•