Genetic Algorithm for CPG controlled Compass Biped
A multi-objective genetic algorithm (GA) was used to obtain parameters for a central pattern generator. The parameters include the controller’s activation frequency and a set of 3 parameters (strength, offset and duration) for each of the torques applied.
DARPA Robotics Challenge: Atlas crawls through mud pit
DARPA Robotics Challenge: Atlas completes Task 2
Simple biped walks with dynamic gait
This video demonstrates a simple implementation of a CPG based controller in a ROS+Gazebo simulation.
The controller uses minimal feedback from the step size to maintain an average walking speed. A passive ankle (spring+damper) was used in conjunction with a toe-off effort. A hip pulse was used to propel the swing leg forward
S.E.A.S.L.: Series Elastic Actuated Scooting Leg
S.E.A.S.L. uses series elastic actuation (SEA) to overcome the friction in the DC motor’s gearbox. Using two encoders (one for the leg and one for the motor) the SEA controller can measure the tension on the system’s elastic elements. This information is used to close a control loop and either apply a desired torque, i.e. maintain a certain angle difference between the leg and the motor, or let the leg swing freely, i.e. keep the angle difference at zero.
S.E.A.S.L. walks using this SEA controller and torque commands coordinated by the biomimetic controller developed in our lab. This controller is based on a Central Pattern Generator (CPG) and takes advantage of the system’s natural dynamics to create an energy efficient, dynamical walk
S.E.A.S.L walking on a treadmill: