Multi-contact Locomotion and Percep- tion on the Humanoid Robot - PowerPoint PPT Presentation

Multi-contact Locomotion and Percep- tion on the Humanoid Robot HRP-2 J. Carpentier C. Quang-Pham A. Del Prete M. Kudruss N. Mansard M. Naveau O. Stasse S. Tonneau Gepetto, LAAS-CNRS, Toulouse, France Int. Conf. on Humanoid Robotics, 10th Workshop on Humanoid Soccer Robots Seoul, Korea, November 3 rd , 2015

Motivations Uncertainity, planning and control Conclusions Presentation overview 1 Motivations Applications Results 2 Uncertainity, planning and control Motion generation Planning complex contact sequences Noise in the contact surfaces Noise in the localization Control and underactuation 3 Conclusions 10th WS on Humanoid Soccer Robots – 2/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Table of Contents 1 Motivations Applications Results 2 Uncertainity, planning and control 3 Conclusions 10th WS on Humanoid Soccer Robots – 3/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Applications Humanoids in Factory like environment 10th WS on Humanoid Soccer Robots – 4/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Results Humanoid robot HRP-2 evolving on stairs [Kudruss, Humanoids 2015] [Carpentier, ICRA 2016 submitted] Previous work [Luo, ICRA 2014] [Vaillant, Humanoids 2014] [Noda, ICRA 2014] 10th WS on Humanoid Soccer Robots – 5/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Results Humanoid robot HRP-2 stepping down [Cuong, IEEE Trans. on Mechatronics 2014] 10th WS on Humanoid Soccer Robots – 6/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Table of Contents 1 Motivations 2 Uncertainity, planning and control Motion generation Planning complex contact sequences Noise in the contact surfaces Noise in the localization Control and underactuation 3 Conclusions 10th WS on Humanoid Soccer Robots – 7/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Motion generation Motion generation: the general problem t  CoM min f ( u ( t ) , v ( t )) Balance (under-actuated part)      g ( u ( t ) , v ( t )) < 0 q ^ GIK A general problem on the time window   h ( u ( t ) , v ( t )) = 0    with u ( t ) the control and v ( t ) the environment model Which v ( t ) for multi- contact control ? 10th WS on Humanoid Soccer Robots – 8/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Planning complex contact sequences Planning complex contact sequences Fast conctact planner from environment CAD (near real-time) [Tonneau, ISRR2015] Evident need of dense mapping as input Preparing force control using robust balance [Del Prete, ICRA 2016 Submitted] 10th WS on Humanoid Soccer Robots – 9/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Noise in the contact surfaces Problems with the environment model Solution Torque control Noise in the contact surfaces Online adaptation to un- known terrain 10th WS on Humanoid Soccer Robots – 10/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Noise in the contact surfaces Problems with the environment model Solution Torque control Noise in the contact surfaces Online adaptation to un- known terrain Torque control for some humanoid robots (HRP-2) is difficult to achieve 10th WS on Humanoid Soccer Robots – 10/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Noise in the contact surfaces Torque control Torque control on a stiff-actuation robot Using end-effector force-torque sensors + IMU + encoders Efficient reconstruction of the motor torques Feedforward on the reconstructed torques (= friction compensation) Feedback on the force sensors (= perfect contact tracking) [Del Prete, IJHR 2015] 10th WS on Humanoid Soccer Robots – 11/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Noise in the localization Problems with the environment model Noise in the localization Rigid robot are good to localize locally SLAM in large environment and use for planning is a challenge In general geometric environment are simple for planning Direct use of geometric models is sometimes preferable Noise due to foot landing and robot Replanning and fast control are necessary 10th WS on Humanoid Soccer Robots – 12/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Control and underactuation Contact and underactuation t CoM  Balance (under-actuated part) c − g ) = � N c m (¨  i = 1 f i   ^ q GIK A general problem on the time window c − g ) = � N c m c × (¨ i = 1 p i × f i    Challenges in Multi-contacts locomotion The general template model includes Quadratic Constraint which can be concave The problem is NP-Hard with c or f as free variables Are the real problems that hard ? Open problem : real-time computation with p i also free variables ? 10th WS on Humanoid Soccer Robots – 13/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Control and underactuation Model-predictive control for 3D locomotion Fast optimal control for central-dynamics pattern generation Near real-time ( 80ms per cycle), ready for MPC Optimize the COM trajectory while keeping the angular momentum low On-going connection with the IMU+force sensor Submitted to ICRA 2016 10th WS on Humanoid Soccer Robots – 14/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Table of Contents 1 Motivations 2 Uncertainity, planning and control 3 Conclusions 10th WS on Humanoid Soccer Robots – 15/16 N. Mansard and O. Stasse

Motivations Uncertainity, planning and control Conclusions Conclusions and Perspectives Conclusions Human environments are still very challenging due to symmetries, lack of textures, occlusion. Including a-priori knowledge helps. Real-time multi-contact based motion generation is difficult Choosing from scratch new contact might be difficult unless candidates are already known. Perspectives Efficient formulation might be found Stochastique approach of control 10th WS on Humanoid Soccer Robots – 16/16 N. Mansard and O. Stasse