Adaptive Compliance in Post-Impact Humanoid Falls Using Preview Control of a Reduce Model

Vincent Samy, Stéphane Caron, Karim Bouyarmane and Abderrahmane Kheddar. To be presented at the 2017 IEEE-RAS International Conference on Humanoid Robots (Humanoids), Birmingham, United Kingdom, November 2017.


We present a novel approach to control a hu-manoid robot in active compliance just after an impact consecutive to a fall. Using linear model predictive control (LMPC), the momentum accumulated by the robot during the falling phase is absorbed, by driving it to zero, until the robot comes to safe rest. The LMPC is written for a reduced center-of-mass model of the robot subject to external contact forces applied on the impact bodies of the robot, each body belonging to an impact limb (arm, leg). Distributing optimally the initial momentum at impact and the total gravity force among all the impacting limbs, we write one LMPC per such limb, each contributing to its own share of the momentum absorption problem. The control vector of each MPC is the contact force applied at the impact body of the limb. We propose a method that allows to encode in a contact polytope both the friction limitations and the actuation torque limits of the impact limb's actuated joints, this polytope models in a synthetic way the linear constraint on the control vector of the LMPC. The approach is validated in full-body dynamics simulation of a humanoid robot falling on a wall.


  title = {{Adaptive Compliance in Post-Impact Humanoid Falls Using Preview Control of a Reduce Model}},
  author = {Samy, Vincent and Caron, St{\'e}phane and Bouyarmane, Karim and Kheddar, Abderrahmane},
  booktitle = {Humanoid Robots, 2017 IEEE-RAS International Conference on},
  year = {2017},
  month = {November},
  note = {to be presented at},
  url = {}
Content on this website is under the CC-BY 4.0 license.