Footstep planning, Future prediction
We are interested in predicting future states of the robot as fast as possible so that to be able to replan in case of strong perturbations. To this aim, we restrict the space of actuation and mimic a simple Linear Inverted Pendulum (LIP). Such simplification allows us to predict future steps easily and we can then use Model Predictive Control (MPC) on footstep planning to control energy of the robot and stability properties. Such framework brings high flexibiltiy and rosutness against various internal and external disturbances as well as strong pushes.
We are basically interested in extending the formulation of this problem so that to handle slopes, rough terrains, stairs and etc.
3LP is a 3D walking model we have recently developed with certain equations and properties to replace our traditional inverted pendulum. Please find movies of 3LP and arXive preprints of papers bellow.
Source codes: https://ponyo.epfl.ch/gitlab/salman.faraji/3LP
The foot-step planner was first applied to the Atlas robot in Gazebo, published in ICRA 2014
It was then applied to Coman robot in webots, published in RSS 2014
Here is also results on transfering the balancing method to the real robot, published in IROS 2015
An extension of the approach to extimate external forces is also submitted to IROS 2016
We are currently working to perform walking as next step.
Feel free to contact us in case you think of possible collaborations.
salman dot faraji at epfl dot ch
A long list of student projects is also available here under humanoids category.